基于改进交叉熵算法的阳极效应预测方法

Anode effect prediction method based on improved cross entropy algorithm

为了准确预测铝电解过程中的阳极效应故障,使用了改进的交叉熵算法(CEM)与支持向量机(SVM)的组合优化算法。针对传统交叉熵算法的无效样本过多、精度不足、可能陷入局部最优解等缺点,提出使用截断高斯分布作为参数采样函数,避免产生非正样本;同时引入“全局精英样本”的概念,保留了历次迭代中的最优样本,优化了参数更新策略;使用快速非支配排序和拥挤距离来计算样本的适应度值,提高了算法的全局搜索性能。实验结果表明,改进后的优化算法可以有效提高阳极效应预测的准确率。

This paper proposed a combined optimization algorithm based on improved cross entropy algorithm(CEM)and support vector machine(SVM),which is used to predict the anode effect accurately during aluminum electrolysis.Aiming at the shortcomings of the traditional CEM,such as too many invalid samples,insufficient precision,and possibly falling into local optimal solutions,it is proposed to use the truncated Gaussian distribution as the parameter sampling function to avoid generating non-positive samples.At the same time,the concept of“global elite sample”is introduced,the optimal samples in previous iterations are retained,and the parameter update strategy is optimized.The fast nondominated sorting and the crowded distance are used to calculate the fitness value of the sample,which improves the global search performance of the algorithm.The experimental results show that the improved optimization algorithm can effectively improve the accuracy of anode effect prediction.