基于GRNN模型的硫化矿石堆氧化自热温度预测

为得到硫化矿石堆氧化自热温度的变化规律，自主设计硫化矿石堆氧化自热模拟试验装置，以含硫量、矿石块度、升温梯度作为试验影响因素，将硫化矿石堆氧化自热温升速率作为试验判定指标，采用L9(34)正交表构造三因素三水平回归正交试验。运用MATLAB建立硫化矿石堆氧化自热温度的GRNN神经网络模型，通过K-折交叉验证优选得到GRNN神经网络的最佳光滑因子σe，并与RBF神经网络模型、灰色神经网络模型预测效果进行对比。结果表明：GRNN神经网络在小样本预测模型中网络逼近能力、收敛速度、算法稳定性等方面具有优势，对硫化矿石堆氧化自热温度的预测精度高，预测误差为3.51%。

Prediction of Oxidation and Self-heating Temperature of Sulfide Ore Heap Based on GRNN Model

Simulation test apparatus of oxidation and self-heating of sulfide ore heap has been designed independently to obtain the change law of the oxidation and self-heating temperature of sulfide ore heap. In the tests,the sulfur content,ore frag-mentation,temperature gradient are taken into account as main influence factors,and the oxidation and self-heating temperature rise rate of sulfide ore heap as a test indicator,L9(34 ) orthogonal table was used to establish the orthogonal regression test of three factors and three levels. GRNN neural network model was established to predict oxidation and self-heating temperature of sulfide ore heap by using MATLAB. K-fold cross validation is applied to GRNN neural network to obtain optimum smoothing factor σe. The RBF neural network model,gray neural network model to predict effects were compared with that of GRNN mod-el predictions. The results show that GRNN neural network has the advantages of network approximation ability, converged speed,and the stability of the algorithm in prediction model of few observations. Prediction accuracy of GRNN model of the oxi-dation and self-heating temperature of sulfide ore heap is high with the prediction error of 3. 51%.