求解广义支持向量机的光滑型算法-控制与决策.pdf

绗?9 鍗 绗? 鏈 鎺 鍒 涓 鍐 绛 2014 骞 4 鏈 Vol. 29 No. 4 Control and Decision Apr. 2014 鏂囩珷缂栧彿: 1001-0920 (2014) 04-0707-06 DOI: 10.13195/j.kzyjc.2012.1891 姹傝В骞夸箟鏀寔鍚戦噺鏈虹殑鍏夋粦鍨嬬畻娉 1a 2 1b 鍊 閾 , 鏉庢案鍒 , 閭佃壇鏉 (1. 杈藉畞宸ョ▼鎶€鏈ぇ瀛. 宸ュ晢绠＄悊瀛﹂櫌锛宐. 绯荤粺宸ョ▼鐮旂┒鎵€锛岃窘瀹佽懌鑺﹀矝 125105; 2. 涓浗涓叅鑳芥簮鑲′唤鏈夐檺鍏徃 閯傚皵澶氭柉鍒嗗叕鍙革紝鍐呰挋鍙 閯傚皵澶氭柉 017200) 鎽 瑕? 鎻愬嚭涓€绉嶆眰瑙ｆ敮鎸佸悜閲忔満(SVMs) 鐨勫厜婊戝瀷绠楁硶. 璇ョ畻娉曞熀浜庡叾瀵瑰伓浼樺寲妯″瀷鐨凨KT 绯荤粺, 鎻愬嚭涓€绫绘柊鐨 鍏夋粦鍑芥暟鏃? 灏嗗叾KKT 绯荤粺閲嶆瀯涓轰竴涓厜婊戞柟绋嬬粍, 骞堕噰鐢ㄥ厜婊戝瀷绠楁硶杩涜姹傝В. 鍦ㄩ€傚綋鐨勬潯浠朵笅, 璇ョ畻娉曟槸鍏ㄥ眬鏀 鏁涘拰灞€閮ㄨ秴绾挎€ф敹鏁涚殑. 澶氫釜绠椾緥琛ㄦ槑璇ョ畻娉曢潪甯告湁鏁? 鍏锋湁骞块様鐨勫簲鐢ㄥ墠鏅? 鍏抽敭璇岤 鏀寔鍚戦噺鏈猴紱鍏夋粦鍑芥暟锛涘厜婊戠畻娉曪紱KKT 绯荤粺 涓浘鍒嗙被鍙耳 TP181 鏂囩尞鏍囧織鐮? A Smoothing type algorithm for solving generalized support vector machines 1a 2 1b NI Tie , LI Yong-li , SHAO Liang-shan (1a. College of Business Administration锛?b. Institute of Systems Engineering锛孡iaoning Technical University 锛孒uludao 125105锛孋hina 锛?. China Coal Energy Company Limited Ordos Branch锛孫rdos 017200锛孋hina 锛嶤orrespondent 锛歂I Tie 锛孍-mail 锛歯itie 4108@) Abstract: A smoothing-type algorithm for solving the generalized support vector machines(SVMs) is proposed. Based on the KKT system of the dual optimization model for SVMs, a new class of smoothing functions is proposed, and the KKT system is reformulated as a system of parameterized smooth equations and solved by using the smoothing type algorithm. Under some reasonable conditions, the proposed algorithm is shown to be globally convergent and locally superlinearly convergent. Numerical examples show that the proposed algorithm is promising and has a bright future of applicat