Deep neural network has emerged as one of the most effective networks for modeling of highly non-linear complex real-time systems. The long-short term memory network (LSTM) which is a one of the variants of recurrent neural network (RNN) has been proposed for the identification of a highly nonlinear Maglev plant. The comparative analysis of its performance is carried out with the functional link artificial neural network- least mean square (FLANN-LMS), FLANN-particle swarm optimization (FLANN-PSO), FLANN-teaching learning based optimization (FLANN-TLBO) and FLANN-black widow optimization (FLANN-BWO) algorithm. The proposed LSTM model is a feed forward neural network trained by a simple iterative method called the ADAM algorithm. The obtained results indicate that the proposed network has better performance than the other competitive networks in terms of the MSE, CPU time and convergence rate. To validate the dominance of the proposed network, a statistical tests, i.e. the Friedman test, is also applied.

中文翻译：

---

基于长期记忆网络的深度学习技术识别实时磁悬浮工厂

深度神经网络已经成为对高度非线性的复杂实时系统进行建模的最有效的网络之一。长期记忆网络（LSTM）是递归神经网络（RNN）的变体之一，已被提议用于识别高度非线性的磁悬浮工厂。使用功能链接人工神经网络-最小均方（FLANN-LMS），FLANN粒子群优化（FLANN-PSO），FLANN示教基于学习的优化（FLANN-TLBO）和FLANN对其性能进行比较分析-黑寡妇优化（FLANN-BWO）算法。提出的LSTM模型是通过称为ADAM算法的简单迭代方法训练的前馈神经网络。获得的结果表明，在MSE，CPU时间和收敛速度方面，所提出的网络具有比其他竞争网络更好的性能。为了验证所提出网络的主导地位，还应用了统计检验，即弗里德曼检验。