Data Mining-Based Model Simplification and Optimization of an Electrical Power Generation System

Abstract

To assess the performance of electrification in an aircraft, multiphysics modeling becomes a good choice for the design of more-electric equipment. However, the high computational cost and huge design space of this complex model lead to difficulties in the optimal design of the electrical power system, thus model simplification is mandatory. For this purpose, this article first proposes a novel model simplification approach based on data mining, and the design of a small electrical power generation system is investigated to demonstrate it. According to the formulated multiphysics model of the system, this article uses the optimal Latin Hypercube-based design of experiment to generate data for the analysis. Based on the generated data, a fusion algorithm integrating multiple feature selection methods is presented to facilitate the dimensionality reduction of the problem's design space. Also, machine learning algorithms are applied to the surrogate model establishment, allowing the reduction of computational time. The investigation of various optimization routines with various multiobjective genetic algorithms shows that the proposed practices improve the system-level optimization efficiency with low computational complexity, ease of search, and high accuracy, which is competitive compared with state-of-the-arts.