基于多值编码GA-BP混合算法的板形板厚综合预测控制

Predictive control of strip flatness and gauge complex control based on hybrid GA-BP algorithm with multi-encoding

为克服BP神经网络存在的收敛速度慢、易于陷入局部极值等不足,提出了可以同时优化BP神经网络的结构和参数的基于多值编码方式的嵌入梯度下降算子的混合遗传算法(GA-BP).并在此基础上针对板形板厚综合系统(AFC-AGC)具有强非线性、强耦合而难以建立精确的数学模型的问题,设计了基于BP网络板形板厚综合预测模型,引入了反馈校正的方法来提高板形板厚控制系统的抗干扰能力.仿真结果表明,该模型可以实现板形板厚的精确控制,为热连轧板形板厚综合控制提供了一个新的有效的方法.

A hybrid genetic algorithm is presented to optimize the structure and parameter of BP （backpropagation） neural networks based on gradient descent of multi-encoding in order to overcome slower convergence and local extremum. Considering the stronger nonlinearity and coupling of strip flatness and gauge complex control, a predictive control model of strip flatness and gauge based on multi-encoding hybrid genetic algorithm （GA） is proposed, and the feedback correction is used to restrain the disturbance of strip flatness and gauge. The simulation results show that the proposed method meets the requirement of precision control and provides a new way for strip flatness and gauge complex control.