安钢FSF电极智能预测建模及应用研究

Intelligent prediction modeling and application research on the Angang FSF electrode

交流电弧炉电极控制系统是一个多变量、非线性、参数时变、复杂强耦合系统,经典的控制策略难以获得优良的性能。为此从安钢FSF电极控制的实际应用出发,提出了1种变结构遗传Elman网络预测建模方法,其中改进的混和遗传算法用来对网络结构和权值及自反馈增益的同步动态寻优。并将基于BP算法的改进Elman网络和本文提出的变结构遗传Elman网络都应用于交流电弧炉的电极模型的辨识中,通过基于安钢现场数据的计算机仿真实验表明:变结构遗传Elman网络克服了因复杂的辨识对象造成的网络辨识结构复杂问题和采用BP算法带来的权值训练缺陷;并具有更好的动态性能,逼近速度快,精度更高等优点。最后,把建立的模型应用于电极控制系统的参数整定上,取得了良好的控制效果,为电极控制提供了理论指导。

The electrode control system of alternating current electrical arc furnace is a multi - variable, nonlinear, parameter-time-varying and complicated strong coupling system and the classic control strategy is hard to acquire the good performance. Proceeding from actual application of FSF electrode control a variable structure Elman neural network prediction model is proposed based on a new hybrid generic algorithm in this paper. This improvized hybrid generic algorithm can simultaneously and dynamically optimize the network structure, the weights and self-feedback gain. The improvized BP algorithm based Elman neural network and the variable structure Elman neural network proposed in this paper are both applied in identification of the electrode model. The simulation experiment based on the on-site data of Anyang Steel indicate that the structure variable Elman neural network overcomes the problem of complex network structure, which is brought by the complexity of electrode control system and limitation of weights by BP algorithm. The proposed method based on a new hybrid generic algorithm is of better identification performance ,better dynamic characteristic, quicker approach speed and better precision. Finally, the model applied in parameter tuning of electrode control system obtains the good control result,thus providing a solid theoretical basis for electrode control.