神经网络控制系统仿真

利用基于神经网络修正误差ＢＰ学习算法的多层网络和间接学习或专门学习的动态逆特性控制方法＾［１］编制的神经网络控制系统的仿真软件（ＳＣＳＢＮＮ），给出了调节时间和最大超调量与神经网络中间层节点数的关系曲线，同时给出了各种学习率和神经元作用函数增益的响应曲线。ＳＣＳＢＮＮ也可用于神经网络非线性控制系统。仿真结果说明神经网络非线性控制系统具有良好的控制性能。

SIMULATION STUDY OF CONTROL SYSTEMS BASED ON NEURAL NETWORK

The programming and debugging of software (SCSBNN) for simulation of neural network control system are implemented. The software (SCSBNN) is based on a feedforward neural network with a modified error-back propagation learning algorithm and indirect learning architecture or specialized learning architecture. Setting time and maximum overshoot versus number of neurons used in the hidden layer are shown. Response curves of an unknown plant controlled by neural networks which use different learning rates and gain of the neurons activation function are also shown. The software (SCSBNN) can be also used for nonlinear systems controlled by neural network. Good controlling capabilities of nonlinear systems by neural networks are demonstrated by the results of simulations.