LF精炼炉电极系统是一个多变量、非线性、时变的复杂系统，传统的PID控制方法使得系统电极位置、电弧长度、电弧电流以及系统功率很难保持在最佳状态下运行，为了改善控制效果需采用先进的控制算法。 论文以某钢厂110t LF精炼炉为研究背景，对其电极调节展开了理论分析及控制方法的研究。首先，论文在介绍了电极构造的基础上建立了其数学模型，并指出了电极系统具有非线性及时变性的特性。其次，论文利用神经网络方法提出了在线递推多步预测模型及在线直接多步预测模型，并通过实际采集的数据验证了在线递推多步预测模型要优于后者。针对此类被控对象，提出了在预测模型基础上采用滚动优化的控制策略，通过仿真与PID控制进行了对比，从仿真结果表明神经网络预测控制具有良好的跟踪特性且超调量小，特别是当电极系统发生变化后能够快速地抑制出现的扰动，说明这种基于神经网络的预测控制可以提高LF精炼炉电极的控制性能。最后，为了模拟实际的运行环境搭建了物理仿真平台，验证了上述方法的有效性。

LF refining furnace electrode system’s characteristics of multiple variable, nonlinearity and changing with time, traditional PID control strategy is not keep electrode position, electrode arc length, electrode arc current, system power work in the optimal state, adopt advance control arithmetic to improve control effort. Papers to a 110 t LF refining furnace for the research background, and its electrode regulator launched a theoretical analysis and control research. First, the papers presented in the electrode structure on the basis of the establishment of the mathematical model, and pointed out that the electrode system with nonlinear time variability characteristics. Secondly, the paper using neural network method of online multi-step forecast model recursive and non-recursive prediction model and the actual data collected through an online verification recursive multi-step prediction model is superior to the latter. Charged against such targets, the forecast made on the basis of a rolling model of the control strategy optimization, simulation and PID control were compared from the simulation results show that the neural network has good predictive control and tracking of a small amount of overshoot, especially When the electrode system is able to change rapidly inhibit the disturbance. This shows that this neural network-based predictive control can improve LF ladle refining furnace electrode performance. Finally, in order to simulate the actual operating environment has set up a physical simulation platform, to verify the validity of the method.