An Approach to Objective Optimization of High-dimensional Data

Abstract

Objective optimization is crucial in industrial production. However, for high-dimensional data, it often needs to take a long time and is full of challenges to get an approximate optimal solution which is very close to the optimum. Therefore, an approach to solve objective optimization problems of high-dimensional data is proposed by combining gradient descent and clustering, with simulated annealing. Considering the optimization problem underneath coal blending in electric field as a background, the proposed approach is compared with the traditional simulated annealing algorithm and that combined with gradient descent. Experiments shows that the proposed approach can improve the average matching degree by 1.94% and 2.14%, and the best matching degree by 0.08% and 0.09%, compared to that based on the traditional simulated annealing algorithm and that combined with superimposed gradient descent respectively. Experiments and analyses also show that the proposed approach can reduce time consumption largely, and has a great improvement in stability, accuracy of the optimal solution as well as the efficiency of solution searching.