基于阻尼最小二乘法的神经网络自校正一步预测控制器

针对非线性控制器设计中遇到的模型结构及模型参数辨识问题，采用多层前馈神经网络去逼近任意的非线性系统，并使用收敛速度快且稳定性好的阻尼最小二乘法在线学习网络的仅植。基于估计的神经网络模型，依据辨识与控制的对偶原则，设计了基于阻尼最小二乘法的一步向前预测控制器。仿真研究表明，这种神经网络自校正控制器不仅具有很好的性能，而且不会产生参数爆发现象。

The Neural Network Self-tuning One-step Predicitive Controller Based on Damped Least Square

The selection of model structure and the identification of model parameters are important problems for nonlinear controller designing. In this paper, it is suggested that multilayer feedforward networks can be used to approximate arbitrary nonlinear systems. The weights of the neural networks are updated by using the damped least square method which is recognized to have faster convergent speed and better stability. Then basing on the estimated neural network model, according to the dual principle between identification and control, the one-step ahead predictive controller was proposed that was technically adapted the damped least square method. The simulation study shows that not noly the neural-net-based self-tuning controller has highly performance, but also it will not produce the phenomena of parameters bursting-off.