Small Data Driven Evolutionary Multi-objective Optimization of Fused Magnesium Furnaces

In most real-world optimization problems, it is very difficult to obtain accurate analytical objective functions derived from process mechanisms. Instead, only approximate objective functions can be built based on sparse historical data. Performance optimization of fused magnesium furnaces is a typical small data-driven optimization problem, where only very limited and noisy data is available. A surrogate-assisted data-driven evolutionary algorithm is proposed in this paper for off-line data-driven optimization of furnaces performance in magnesia production. The multiobjective evolutionary algorithm is assisted by Gaussian process models to search for Pareto optimal solutions. To generate new data samples in surrogate management, a low-order polynomial model is constructed as an approximate mechanism model that can be treated as the real fitness function. To verify the effectiveness of the proposed Gaussian process assisted evolutionary algorithm, it is first tested on nine benchmark problems in comparison with a popular multi-objective evolutionary algorithm and a surrogate-assisted evolutionary algorithm. The algorithm is then applied to a real-world fused magnesium furnaces optimization problem.