Operating performance assessment methods are performed to evaluate the operating performance of industrial processes so as to provide helpful guidance for safe and high-efficient operations. The rolling processes are always featured with plant-wide characteristics and equipped with hierarchical information systems, while traditional assessment methods are centralized, which, thus, are less efficient. To solve this problem, this paper proposes a method that integrates hierarchical and block information. Firstly, a hierarchical strategy is adopted, and the entire process are partitioned into several levels and sub-blocks such that the interpretability of the assessment results can be improved. Then, in order to extract comprehensive operating state features, a combination of kernel principal component analysis and t-distributed stochastic neighbor embedding-based method is proposed, which can capture both global and local features in a parallel manner. In addition, to overcome the hard decision-making results caused by support vector machines(SVMs), this paper transfers the results of the SVM to calculate the posterior probability, and makes use of the D-S evidence theory to fuse the decision-making process. Finally, the proposed method is applied to a real hot rolling mill process, the results show that it improves at least 18% assessment rate compared with several traditional methods.