full2010.pdf - page 454

416
‹¹ŠÅ—o¡¥µ¥µ¤Ä®o ‡ªµ¤Á¦nŠÁ o µ®µ gbest ¤¸‡nµ¤µ„„ªnµ pbest Á¡ºÉ°Ä®o¨¼nÁ oµ¼n‡Îµ˜°Á¦Èª¥·ÉŠ ¹Êœ) Î µ®¦´Ã‡¦Š nµ¥ž¦³µš
Áš¸¥¤»—šoµ¥ ¤¸ nªœž¦³„°Á®¤º°œ„´  5 ǦŠ nµ¥ž¦³µšÁš¸¥¤ o µŠ˜oœ—´Š„¨nµª oµŠ˜o œ ˜n‹³˜nµŠ„´œÁ¡¸¥Š Output š¸É
ŗo œ´Êœ‹³Å—o°°„¤µÁž}œ 5 Output Ž¹ÉŠ‹³™¼„čoÁž}œ Weight Ĝ„µ¦‡Îµœª–¦µ‡µžd —Äœ ´Êœ»—šoµ¥˜n°Åž
3. Fuzzy Back Propagation Neural Network čoǦŠ nµ¥ž¦³µšÁš¸ ¥¤š´ÊŠ®¤— 6 ǦŠ nµ¥ž¦³µšÁš¸ ¥¤
ÁnœÁ—¸ ¥ª„´ Fuzzy Particle Swarm with IDDM Neural Network ‹µ„„µ¦š—°®»o œš´ÊŠ 50 ®»oœ ×¥Áž¦¸¥Áš¸¥
¦³®ªn µŠ 1, 15, 30 ¡ªnµ ®»oœnªœÄ®n ¤¸‡nµÁŒ¨¸É¥ °Š MAPE š¸Éŗo ‹µ„ 15 Nodes œ´Êœœo°¥š¸É »— ˜nÁœºÉ°Š‹µ„‹Îµœªœ
Hidden Layer š¸ÉÁ®¤µ³¤œ´Ê œ‹ÎµÁž}œš¸É ‹³˜o°Šš—°˜µ¤®»oœÂ˜n¨³˜´ª ץĜǦŠŠµœœ¸Ê ŗo¤»nŠÁœoœÅžš¸É„µ¦®µ
‹Îµœªœ °Š Hidden Layer š¸ÉÁ®¤µ³¤ ‹¹ŠÅ—oÁ¨º°„Á°µ 15 Node ¤µÄo Ĝ„µ¦ train 500,000 ¦° ץčo ÁŠºÉ°œÅ
šµŠ—o µœ °ŠÁª¨µÁ¡ºÉ°Ä®o Á®¤µ³¤„´„µ¦œÎ µÅžÄoĜ Web Application š¸É˜o°Š„µ¦‡ªµ¤¦ª—Á¦ÈªÁž}œ®¨´„ 宦´
ǦŠ n µ¥ž¦³µšÁš¸¥¤»—šo µ¥ ¤¸ nªœž¦³„°Á®¤º°œ„´„´ 5 ǦŠ n µ¥ž¦³µšÁš¸¥¤š¸É„¨nµª¤µÂ¨oª oµŠ˜o œ ‹³
˜nµŠ„´œÁ¡¸¥Š Output š¸Éŗo œ´Êœ‹³°°„¤µÁž} œ 5 Output Ž¹ÉŠ‹³™¼ „čoÁž}œ Weight Ĝ„µ¦‡Îµœª–¦µ‡µžd— ´Ê œ»—šoµ¥
˜n°Åž
Ÿ¨„µ¦«¹„¬µ
„µ¦š—°˜´ªÂ¡¥µ„¦–r®»o œœ¸Êčoš—°„´ ®»oœ—´Š˜µ¦µŠš¸É 1 š¸É¤¸„µ¦ŽºÊ° µ¥Äœ˜¨µ—®¨´„š¦´¡¥r®nŠ
ž¦³Áš«Åš¥ ×¥œÎµ o°¤¼¨¤µÂn Š°°„Áž}œ 2 »— ‡º°» —š¸Éčo 宦´Á¦¸¥œ¦¼o¨³»—š¸Éčo 宦´š—° ¨³Á¤ºÉ°Á¦È ‹
·Êœ„¦³ªœ„µ¦Á¦¸ ¥œ¦¼o¨o ª œÎµ˜´ ªÂš¸É ŗo¤µš—°¡¥µ„¦–r „´ o°¤¼¨»—š—° ץ˜n ¨³®»oœÅ—oš—°
¡¥µ„¦–r 1 ¨³ 7 ª´ œ¨nªŠ®œoµ nŠÁž} œš—°¦µ¥že ¨oª®µ‡nµÁŒ¨¸É¥ °Š‡nµ°´˜¦µ„µ¦šÎµœµ¥™¼„¨³‡nµ‡ªµ¤Á¦Èªš¸É čo
Ĝ„µ¦Á¦¸¥œ¦¼o ‹µ„„µ¦š—°Ã—¥Äo £µª³Âª—¨o°¤˜µ¤š¸Éŗo„¨nµª¤µÂ¨oª ŗoŸ¨¨´ ¡›r—´Šœ¸Ê
1. ‡ªµ¤™¼„˜o°ŠÄœ„µ¦šÎµœµ¥
宦´„µ¦š—¨°Š‡¦´ÊŠœ¸Êčo‡nµ Mean Average Percentage Error Ž¹ÉŠ¤¸ ¼˜¦—´Šœ¸Ê
×¥„ε®œ—Ä®o Price1 ‡º°¦µ‡µ®»oœª´œ™´—Åž, Price 2 ‡º°¦µ‡µ®»oœš¸ÉšÎµœµ¥ ץčo ¤„µ¦—´Š„¨nµªÄœ„µ¦ª´—
‡ªµ¤™¼„˜o °ŠÃ—¥¥¹—‡ªµ¤µ¤µ¦™Äœ„µ¦ Test ªnµ¤¸ž¦³·š›·£µ¡¤µ„œo °¥Á¡¸¥ŠÄ—Áž} œ˜´ª¸Êª´— Ÿ¨¨´¡›r ‹³Â—Š‡nµ
Error ¨³‡nµ MAPE ŗo—´Š˜µ¦µŠš¸É 3 ¨³ 4 ˜µ¤¨Îµ—´
50
1
1 (| Pr 1 Pr 2 |*100) / Pr 1
50 i
ice ice ice ¦