基于近邻转移约束规则的非确定工业过渡过程的模态识别方法

Modal identification method of uncertain industrial transition process based on nearest neighbor transfer constraint rule

受外界环境干扰、生产计划变更和调控方案调整的影响,一些工业过程在不同稳定工况间切换时具有较大的不确定性,给过渡模态识别带来了很大的困难,影响了后续的过程监测和调控优化。结合近邻思想和模态转移约束思想,提出了基于近邻转移约束规则的过渡模态识别方法。该方法使用变量贡献表征模态信息进行历史工况的过渡模态划分;在此基础上,建立了近邻转移约束规则,对过渡模态进行两步识别:第一步引入移动窗口,采用近邻思想计算待识别窗口工况在各过渡模态簇中的近邻贡献距离,识别候选过渡模态;第二步构建历史非稳定工况的过渡模态转移概率矩阵作为权值矩阵,对待识别窗口工况在各候选过渡模态中的近邻贡献距离进行加权、比较,精准定位待识别窗口工况所属模态。以矿渣粉磨生产作为案例进行了应用验证,与其他方法相比,所提方法的模态识别滞后时间更短,识别准确率提高到98.10%。

Affected by external environment interference,production plan change and regulation scheme adjustment,some industrial processes have great uncertainty when switching between different stable conditions,which brings great difficulties to transition mode identification and affects the subsequent process monitoring and regulation optimization.Therefore,combined the idea of nearest neighbor and mode transfer constraint,a transitional modal identification method was proposed based on the nearest neighbor transfer constraint rules.The transition mode of historical working condition was divided by using the modal information represented by the variable contribution.On this basis,the nearest neighbor transfer constraint rule was established to identify the transition modes in two steps:the first step was to introduce the moving window,and the nearest neighbor contribution distance of the window condition to be identified in each transition mode cluster was calculated by using the nearest neighbor idea to identify the candidate transition modes;the second step was to build the transition mode transition probability matrix of historical unstable conditions as the weight matrix,weight and compare the nearest neighbor contribution distance of the window conditions to be identified in each candidate transition mode,and accurately locate the mode of the window conditions to be identified.Taking the actual production of slag grinding system as an example,the modal identification lag time of the proposed method was shorter than that of other methods,and the identification accuracy was higher,reaching 98.10%.The application effect had been verified.