基于MATLAB的BP神经网络对电弧炉炼钢的终点预报

End-point Forecasting of EAF Steelmaking by BP Neural Network Based on MATLAB

针对电弧炉炼钢终点的重要性,建立了基于MATLAB的BP神经网络对电弧炉炼钢终点碳含量及温度预报的模型,运用传统试凑法和单向调节法相结合的方法,可快速地找到最佳隐含层节点数,节省网络训练时间,具有实用性。对比预报线性回归相关系数可知,在0.01显著水平下终点碳含量和终点温度的相关系数均通过显著性检验;终点碳含量和温度的预报值和实测值的整体相关系数分别为0.899和0.820,均高于0.75,表明预测值与实测值的相关性十分显著,该网络模型具有极好的预报性;终点碳含量和温度在误差范围为±0.02%和±10℃内的预报命中率达到91.2%和94%,表明运用神经网络对电弧炉炼钢终点进行预报是有效的、可行的。

Aimed at the complexity, nonlinearity and uncertainty of the process and the importance of the end-point about electric arc furnace steelmaking, a model to forecast the end-point carbon content and temperature of EAF steelmaking by neural network was established based on MATLAB, and the best number of hidden layer nodes can be found quickly combining the traditional method of trial and error with one-way adjustment method. By contrast the linear regression correlation coefficient shows that both the correlation coefficient of the end-point carbon content and temperature are through the test of significance under 0.001 significant level; the whole correlation coefficient of the forecast value and the actual values of the end-point carbon content and temperature is 0.899 and 0.820, respectively, it shows that the correlation between the forecast value and the actual value is very significant for they are both higher than 0.75 and the neural network model has a well forecasting performance. The forecast ratio of the end-point carbon content and temperature is 91.2% and 94% when the error is in the range of ±0.02% and ±10 ℃, respectively. So using neural network to forecast the end-point of electric arc furnace steelmaking is effective and feasible.