步进梁加热炉神经网络模型预测控制

Neural Network Model Predictive Control for Steel Rolled Reheating Furnace

针对步进梁式连续加热炉燃烧过程控制的温度分布非线性和滞后性,提出了一种基于非线性优化技术的神经网络模型预测控制算法。神经网络具有强大的自学习和非线性映射能力,把神经网络预测模型和非线性优化器整合为一个温度控制器,通过神经网络预测模型描述温度控制对象的动态行为,根据加热炉当前温度和出钢温度预测未来时刻的温度输出值,实现加热炉温度控制。实验结果表明,通过对网络模型进行大样本训练和对模型预测控制参数的优化,加热炉温度控制系统能快速达到控制要求,具有良好的抗干扰能力和温度跟随性能。

Neural network model predictive control (NN-MPC) based on NLO is presented according to the characteristics of nonlinearity and time delay of reheating furnace temperature system. Benefited from the powerful self-learning and nonlinear mapping of neural network, the temperature controller, which combined NN-PM with NLO, achieves the control of reheating furnace temperature, namely, it describes dynamic performance by NN-PM and predicts the future output based on current temperature and set-point. Experimental results demonstrate the good control performance of temperature controller such as robust and temperature tracking through training with great amounts of input/output data and optimizing MFC parameters by using NN-MPC.