基于同步优化的代价敏感支持向量机优良类操作模式识别

Excellent operational pattern recognition based on simultaneously optimizing cost-sensitive support vector machine

针对氧化铝蒸发过程操作模式集中类别不平衡和噪声特征问题,提出基于同步优化的代价敏感支持向量机操作模式识别方法。对氧化铝蒸发过程机理进行分析,该过程的输入条件、操作参数和状态参数被选为原始操作模式,利用离散的粒子群算法优化操作模式的特征集,选择最优特征子集作为最终的操作模式;同时利用连续的粒子群算法优化代价敏感支持向量机的核参数和误分类代价参数,自动搜索和确定最优的核参数和误分类代价参数。工业应用结果表明,与粒子群优化操作模式特征子集或粒子群优化核参数和误分类代价参数相比,所提出的方法优良类操作模式识别高,误分类代价低。

Aiming at class-imbalanced operational pattern recognition and noise features for alumina evaporation process （AEP）, a simultaneously optimizing cost-sensitive support vector machine （CSVM） is proposed in this paper. Studying the mechanism of AEP, input conditions, operating parameters and state parameters are selected as original operational pattern. The feature set of original operational pattern are optimized by the binary particle swarm optimization and the optimal feature subset is selected as the operational pattern. Meanwhile, the sigma of Gaussian kernel and miselassification cost parameters for CSVM are optimized by the linear weight diminishing particle swarm optimization. The proposed method is applied on the operational pattern optimization of AEP. Experimental results illustrate that the proposedmethod increases excellent operational pattern recogn{tion rates and reduces misclassification costs.