摘要
提出了一种基于神经网络模型的非线性多步预测控制策略。预测器和控制器由一个BP网络构成。在整个过程中,首先利用一个BP网络构造一个非线性多步预测模型,根据被控对象输出与网络实际输出之间的误差采用改进的BP算法修改网络权值,以逐步建立合理的多步预测模型。然后,根据网络的多步预测输出序列与设定值序列的偏差构造性能指标函数,根据性能指标函数采用自适应变步长梯度法修改控制律。仿真结果表明了该策略的有效性。
The paper presents a kind of nonlinear multi-step predictive control strategy based on neural network model.The predictor and the controller are realized by a network. During the whole process, we at first build a nonlinear multi-predictive model by adopting BP network and use improved BP algorithm to revise the values of the weights of the network for creating a reasonable multi-step output prediction model by the bias between the BPN's productions and the plant's outputs. Then the target function is built by the bias between the outputs sequence of the multi-step prediction neural network model and the setpoint sequence of the system. We adopt adaptive variable-step gradient algorithm to revise the control law based on the target function . The simulation experiment has proved the validity of the algorithm.
出处
《计算机仿真》
CSCD
2004年第12期152-154,共3页
Computer Simulation
关键词
神经网络
梯度法
预测控制
Neural network
Gradient algorithm
Predictive control