Optimization of Transformer Windings Based on FEA‐XGBoost and NSGA‐III Algorithm

• Published in: Journal of Electrical and Computer Engineering Vol. 2024; no. 1
• Main Authors: Bai-di, Shi, Yong-feng, Jiang, Jing-yu, Shang, Ye-feng, Bao, Bing-yan, Chen, Ke, Yang
• Format: Journal Article
• Language: English
• Published: New York John Wiley & Sons, Inc 2024; Wiley
• Subjects: Accuracy; Algorithms; Analysis; Coils (windings); Cost analysis; Design; Design analysis; Design optimization; Design parameters; Electric transformers; Energy distribution; Energy transmission; Engineering; Environmental protection; Finite element method; Genetic algorithms; Heat; Heuristic; Impedance; Leakage; Load; Losses; Machine learning; Manufacturing; Manufacturing costs; Methods; Multiple objective analysis; Optimization; Optimization models; Power; Production costs; Short circuit currents; Transformers
• ISSN: 2090-0147; 2090-0155; 2090-0155
• DOI: 10.1155/2024/5514678

Power transformers are indispensable components for energy transmission and voltage regulation. Since the leakage impedance affects the short‐circuit current, magnetic leakage distribution, and manufacturing cost, the accurate calculation of transformer impedance is vital for transformer design. Generally, leakage impedance is mainly decided by the design parameters of the windings, both analytically and numerically. In most given literature studies, the leakage impedance was optimized and analyzed by adjusting the design parameters of the windings without considering the consequent influence on transformer loss and manufacturing costs. A multiobjective optimization model considering the leakage impedance, manufacturing cost, and operating loss of the windings is presented in this paper. The eXtreme Gradient Boosting (XGBoost) model is built using 2048 finite‐element analysis (FEA) samples and utilized as a leakage impedance predictor. XGBoost shows better accuracy and consumes much less time compared to analytical and FEA methods. Subsequently, the presented multiobjective model is optimized using multiobjective algorithms and the NSGA‐III shows the best performance among the NSGA‐II, MOPSO, and MODE. The results show that the leakage impedance is closer to the required value. Besides, the winding manufacturing cost and loss are meanly decreased by 4.7% and 4.1%, respectively, in the engineering case.