运用神经网络预报烧结终点

Prediction of Sintering BTP Based on Artificial Neural Network

针对现场烧结终点控制复杂与难度大的实际,开发了神经网路预报系统。预报系统采用4层前向神经网络,进行多因素输入建模,输出采用具有极值特性的二次曲线计算的烧结终点与实际最高废气温度,预报烧结终点与最高废气温度,为现场终点控制的最新可行方法。网络结构设计先进合理、精度高、泛化能力强,训练方差为0.00001814,用训练样本集测试输出,烧结终点绝对平均误差为0.04,终点废气温度绝对平均误差为4.57℃。采用训练后网络预报,烧结终点(风箱号)绝对误差最大仅为0.09,终点废气温度绝对误差最大为3.57℃,命中率100%。用预报结果有针对性调节烧结参数可收到明显效果。

Aim at actual complicacy and difficulty of BTP （Burning Through Point） control,a prediction system of neural network has been developed. The model of 4 layers feedforward neural network with many factors has been set up for forecasting BTP and the highest exhaust gas temperature, that network output is constituted by compu- ting BTP and the highest exhaust gas temperature with extremum characteristic of conic, what is new method of locale BTP to control. The network possess advanced reasonable construction designs, high accuracy and strong generalization ability. The network training sum of squared error is 0. 00001814. To test the output with the training sample set, the absolute average error of BTP is 0.04,and the highest exhaust gas temperature is 4.57X2 at BPT bellows. The maximal absolute error of BTP（NO. of bellows） is 0.09 when forecasting after neural network training, it is 3. 57X2 for the highest exhaust gas temperature at BPT bellows, the rate in accuracy area of forecast is 100M. A well result will be obtained by adjusting sintering parameter with intention.