Shuqing Li

In recent years, the multi-criteria optimization classifier (MCOC) approach based on optimization theory has gained considerable research attentions and practical applications. Compared to traditional support vector classifier (SVC), MCOC does not strictly conform to a singular optimization criterion. Instead, they incorporate multiple, often inconsistent, objectives into their decision-making process, thereby providing a more holistic approach to classification. To reduce the effect of redundant features, select important ones and enhance the interpretability of the model, regularization sparsity methods introduce sparsity constraints into the optimization function. This promotes the reduction of many weights of features to zero or near zero, thus achieving the goal of feature reduction. The [ ]-norm regularization is considered the optimal regularization function due to its direct control over model complexity. However, the nonconvexity and discontinuity of [ ]-norm make the problem particularly challenging to solve. Consequently, researchers have been striving to find the optimal convex approximation of the [ ]-norm to facilitate more effective optimization. Multiple kernel learning (MKL) utilizes optimal combinations of kernel functions to effectively map data into high-dimensional spaces, achieving linear separability in nonlinear datasets. However, it does impact the selection of important features to some extent. A novel sparse multi-kernel multi-criteria stochastic optimization classifier (SMMSOC) is proposed in this paper. The model incorporates a convex approximation of [ ]-norm regularization and a redefined multi-kernel matrix into a nonlinear MCOC framework. The optimization algorithm is implemented using the Nesterov accelerated gradient (NAG) algorithm integrated with historical gradient averaging based on mini-batch data sets. Experimental results demonstrate that the SMMSOC model enhances both classification accuracy and model sparsity.

This paper firstly proposes an assessment index system for source-side integrated energy systems; secondly, the indexes are assigned weights through a combination of subjective and objective weighting methods by the set-value iterative method and the inverse entropy weighting method, and are evaluated using the object topology model; finally, an evaluation analysis is carried out in relation to a proposed source-side integrated energy system to verify the rationality of the proposed assessment index system and the effectiveness of the assessment method