基于FOA-GRNN模型的转炉炼钢终点预报

Endpoint prediction of basic oxygen furnace steelmaking based on FOA-GRNN model

目前广泛采用的RBF神经网络具有训练时间长与训练困难等缺陷.本研究结合实际生产数据,建立了FOA-GRNN神经网络预报模型,并对转炉终点温度与碳质量分数进行预报.结果表明:与RBF神经网络相比,FOA-GRNN神经网络可以有效提高命中率并满足实际生产要求.当碳质量分数绝对误差小于±0. 03%时,FOA-GRNN神经网络预报命中率可由91%提高至94%;当温度绝对误差小于±15℃时,预报命中率可由89%提高至97%.同时,FOA-GRNN神经网络训练时间在RBF神经网络基础上分别降低了42. 22%与37. 08%,预报结果与实测值的均方差也有一定的降低,故可为现场生产提供重要的参考.

The widely used RBF neural network nowadays has shortcomings of long training time and difficult training.Based on the production data,a FOA-GRNN model was established to predict the end-point temperature and carbon content in the present paper.The results showed that the hit rates of the FOA-GRNN model can meet requirement of the production and are higher than that of RBF model.When the absolute error of predicted carbon content is within±0.03%,the accuracy of the model increases from 91%to 94%.When the absolute error of predicted temperature is within±15℃,accuracy of the model increases from 89%to 97%.Meanwhile,the training time decreases 42.22%and 37.08%and the mean square errors also decrease.So that it can provide an important reference for practical applications.