Zhang Xiaoyu, Wang Degang, Yang Bin, Ota Kaoru, Dong Mianxiong, Li Hongxing
IEEE Trans Neural Netw Learn Syst. 2024 Mar;35(3):3713-3724. doi: 10.1109/TNNLS.2022.3196402. Epub 2024 Feb 29.
Time delay has always been one of the main factors affecting the application performance of neural network (NN) systems, and dynamic performance research of NNs with time delays has been the focus of many scholars in recent years. This article enquires into the exponentially synchronous problem of switched delayed NNs with time delay in the leakage term. Adopting an unusual form from a common switched system, the switching modes of the switched delayed NNs system in this article are dependent on time delays. In the first place, the master, slave, and error NNs models are reconstructed into the switched form by introducing the switched delay idea. Then with the help of the admissible edge-dependent average dwell time (AED-ADT) method and delay-dependent switching adjustment indicators, a novel set of generalized delay-mode-dependent multiple Lyapunov-Krasovskii functionals (MLKFs) is built for analyzing the cases where a state-feedback controller exists and does not exist in the model, and where parts of LKFs may increase during the period when the corresponding subsystems are activated. For these cases, several effective exponential synchronization criteria and switching laws are presented accordingly. At last, the verification of the theoretical results is shown through a few examples.
时间延迟一直是影响神经网络（NN）系统应用性能的主要因素之一，近年来，具有时间延迟的神经网络的动态性能研究一直是众多学者关注的焦点。本文研究了泄漏项具有时间延迟的切换延迟神经网络的指数同步问题。本文采用了一种不同于普通切换系统的形式，即切换延迟神经网络系统的切换模式依赖于时间延迟。首先，通过引入切换延迟思想，将主、从和误差神经网络模型重构为切换形式。然后，借助允许的边依赖平均驻留时间（AED-ADT）方法和延迟依赖切换调整指标，构建了一组新颖的广义延迟模式依赖多重Lyapunov-Krasovskii泛函（MLKF），用于分析模型中存在和不存在状态反馈控制器的情况，以及在相应子系统激活期间部分Lyapunov-Krasovskii泛函可能增加的情况。针对这些情况，相应地给出了几个有效的指数同步准则和切换律。最后，通过几个例子对理论结果进行了验证。
