ABSTRACT

This paper introduces a nonlinear adaptive controller of unknown nonlinear dynamical systems based on the approximate models using a multi-layer perceptron neural network. The proposal of this study is to employ the structure of the Multi-Layer Perceptron (MLP) model into the NARMA-L2 structure in order to construct a hybrid neural structure that can be used as an identifier model and a nonlinear controller for the MIMO nonlinear systems. The big advantage of the proposed control system is that it doesn’t require previous knowledge of the model. Our ultimate goal is to determine the control input using only the values of the input and output. The developed NARMA-L2 neural network model is tuned for its weights employing the backpropagation optimizer algorithm. Nonlinear autoregressive-moving average-L2 (NARMA-L2) neural network controller, based on the inputs and outputs from the nonlinear model, is designed to perform control action on the nonlinear for the attitude control of unmanned aerial vehicles (UAVs) model. Once the system has been modeled efficiently and accurately, the proposed controller is designed by rearranging the generalized submodels. The controller performance is evaluated by simulation conducted on a quadcopter MIMO system, which is characterized by a nonlinear and dynamic behavior. The obtained results show that the NARMA-L2-based neural network achieved good performances in modeling and control.

摘要

本文介绍了一种基于使用多层感知器神经网络的近似模型的未知非线性动力系统的非线性自适应控制器。本研究的建议是将多层感知器（MLP）模型的结构应用到 NARMA-L2 结构中，以构建一种混合神经结构，该结构可用作 MIMO 非线性的标识符模型和非线性控制器系统。所提出的控制系统的一大优点是它不需要模型的先前知识。我们的最终目标是仅使用输入和输出的值来确定控制输入。开发的 NARMA-L2 神经网络模型使用反向传播优化器算法对其权重进行调整。非线性自回归-移动平均-L2 (NARMA-L2) 神经网络控制器，基于非线性模型的输入和输出，旨在对无人机（UAV）模型的姿态控制的非线性执行控制动作。一旦系统被有效且准确地建模，所提出的控制器就通过重新排列广义子模型来设计。通过在四轴飞行器 MIMO 系统上进行仿真来评估控制器性能，该系统具有非线性和动态行为的特点。所得结果表明，基于NARMA-L2的神经网络在建模和控制方面取得了良好的性能。所提出的控制器是通过重新排列广义子模型来设计的。通过在四轴飞行器 MIMO 系统上进行仿真来评估控制器性能，该系统具有非线性和动态行为的特点。所得结果表明，基于NARMA-L2的神经网络在建模和控制方面取得了良好的性能。所提出的控制器是通过重新排列广义子模型来设计的。通过在四轴飞行器 MIMO 系统上进行仿真来评估控制器性能，该系统具有非线性和动态行为的特点。所得结果表明，基于NARMA-L2的神经网络在建模和控制方面取得了良好的性能。