TW201118764A - Ship dynamic mode of operation of early warning systems for windows - Google Patents
Ship dynamic mode of operation of early warning systems for windows Download PDFInfo
- Publication number
- TW201118764A TW201118764A TW98140565A TW98140565A TW201118764A TW 201118764 A TW201118764 A TW 201118764A TW 98140565 A TW98140565 A TW 98140565A TW 98140565 A TW98140565 A TW 98140565A TW 201118764 A TW201118764 A TW 201118764A
- Authority
- TW
- Taiwan
- Prior art keywords
- ship
- typhoon
- early warning
- wind
- learning
- Prior art date
Links
Landscapes
- Traffic Control Systems (AREA)
Abstract
Description
201118764 六、發明說明: 【發明所屬之技術領域】 本發明係關於一種利用類神經網路所建立的預警模 式,尤指一種可預測船舶動態並具有圖形使用者介面 (Graphical User Interface,GUI)的預警模式。 【先前技術】 台灣濱臨太平洋,位處易受颱風或異常波浪侵襲之地 區,而颱風所引起的巨浪為破壞港灣的重要因素,其亦困 擾港内船舶作業與碇靠之管理。 …因此’港内之穩靜情況關係著船隻碗泊與貨物裝卸作 業管理之重要。鑑於料或異常波浪料灣設施及船隻碗 泊的重要性’需建立一個可即時性反應的船舶動態管理系 統,俾提供料單位船隻碗泊、裝卸作業及離港疏 考依據。 【發明内容】201118764 VI. Description of the Invention: [Technical Field] The present invention relates to an early warning mode established by using a neural network, in particular, a predictable ship dynamic and having a graphical user interface (GUI) Early warning mode. [Prior Art] Taiwan is located in the Pacific Ocean, where it is vulnerable to typhoons or unusual waves. The huge waves caused by typhoons are an important factor in destroying the harbor. It also plagues the management of ship operations and dependencies in the port. ...therefore, the stability of the port is related to the management of vessel bowling and cargo handling operations. In view of the importance of material or abnormal wave material bay facilities and vessel bowling, it is necessary to establish a ship dynamic management system that can respond immediately, and provide the basis for vessel parking, loading and unloading operations and departure inspection. [Summary of the Invention]
本發明的目的之係提供一種船舶動態預警系統視 函化操作模式,用以預測船隻異動情形。 本發明揭露-種船舶動態預警系統視窗化操作模 ^ 貝'料整合模、组 '一類神經網路(Artificial Neural 資JTHNN)處理模M、"'儲存模組、以及一顯示模組。 貝料整&模組,供一传用去於λ 使用者輸入一參數,並將該參數進行 堂❼,類神經網路處理模 ^ ^ ^ 偶、·且將參數進仃運算,儲存模組, 存㈣神經網路處理模叙之―資料;顯示模組,顯示儲 201118764 存模組所儲存之該資料。其中,船舶動態預警系統視窗化 操作模式係用以預測一船隻異動指數(Index For Ship Escape , ISE)〇 【實施方式】 在本發明一實施例中,由花蓮港務局持續觀察花蓮港 港池受颱風影響之狀況,影響港池之程度概取決於颱風之 路徑、強度、暴風圈範圍及其行進速度等因素,本實施例 就台灣東部海域民國85年至94年共82場颱風對花蓮港 港池狀況之影響歸納進行七種路徑分類,此七種颱風路徑 分類法示如表1。 表1花蓮港務局颱風路徑歸類表It is an object of the present invention to provide a visualized operational mode of a ship dynamic early warning system for predicting a ship's changing condition. The invention discloses a windowing operation mode of a ship dynamic early warning system, a material integration module, a group of neural networks (JTHNN) processing module M, a storage module, and a display module. The shell material & module, for a pass to the user input a parameter, and the parameter is performed, the neural network processing module ^ ^ ^ even, and the parameters into the operation, storage mode Group, save (4) neural network processing model - data; display module, display the data stored in the 201118764 storage module. The window operation mode of the ship dynamic early warning system is used to predict an index of Ship Escape (ISE). [Embodiment] In an embodiment of the present invention, the Hualien Port Authority continuously observes the Hualien Port Basin. The extent of the impact of the typhoon, the extent of the impact on the harbor basin depends on the path of the typhoon, the intensity, the extent of the storm circle and its speed of travel. In this example, a total of 82 typhoons from the 85-94 period in the eastern Taiwanese waters to the Hualien Port Basin The impact of the situation is summarized in seven path classifications. The seven typhoon path classifications are shown in Table 1. Table 1 Classification table of typhoon path of Hualien Port Authority
颱風路徑 颱風名稱 個數 1 CAM(凱姆)、NOGURI(諾古力)、NANGKA(南 卡)、CONSON (康森) 4 2 BOPHA(寶發)、NARI(納莉) 2 3 BETH(貝絲)、ERNIE(爾尼)、BABS(芭比絲)、 FAITH(費絲)、IMBUDO(尹士都)、 KROVANH(柯羅旺) 6 4 AMBER(安珀)、YANNI(楊妮)、BILIS(碧利 斯)、 LEKIMA (利奇馬)、TORAJI(桃芝)、 TRAMI (潭美)、MORAKOT (莫拉克)、 MINDULLE (敏督利)、NOCK_TEN(納 坦)、HAITANG (海棠)、LONGWANG (龍王)、 TALIM (泰利) 12 5 HERB(賀伯)、WINNIE(溫妮)、ZEB(瑞伯)、 KAI_TAK(啟德)、PRAPIROON(巴比舍)、YAGI (雅吉)、RAMMASUN (雷馬遜)、KUJIRA (柯吉拉)、MAEMI (梅米)、MEARI (米雷)、 MEGI (梅姬)、RANANIM (蘭寧)、SONGDA (桑 達)、TOKAGE (陶卡基)、KHANUN (卡努) 15 6 GLORIA(葛樂禮)、SALLY(莎莉)、DAN(丹 恩)、MAGGIE(瑪姬)、SAM(山姆)、CHEBI(奇 比)、UTOR(尤特)、DUJUAN(杜鵑)、MELOR(米 勒)、KOMPASU(康柏斯)、OTTO(奥托) 11 201118764 AERE (艾利)、DALE(戴兒)、VIOLET(魏萊 特)、ZANE(薩恩)、ISA(麗莎)、IVAN(艾文)、 LEVI(里維)、OPAL(歐珀)、PETER(彼得)、 ROSIE(羅西)、TINA(蒂納)、VICK丨(維琪)、 BART(巴特)、GLORIA(葛樂禮)、KATE(凯 特)、OLGA(歐佳)、DAMREY(丹瑞)、 KIROGI(奇洛基)、XANGSANE(象神)、 CIMARON(西馬隆)、HAIYAN(海燕)、ETAU(艾 陶)、LINFA(蓮花)、LUPIT(盧碧)、 SOUDELOR(蘇迪勒)、CHABA(佳芭)、 MA—ON(馬鞍)、NIDA(妮妲)、SUDAL(舒達)、 NABI(娜比)、NESAT(尼莎)、SONCA(桑卡)Typhoon path Typhoon name number 1 CAM (Kem), NOGURI (Norocal), NANGKA (South Carolina), CONSON (Conson) 4 2 BOPHA (Baofa), NARI (Nari) 2 3 BETH (Beth ), ERNIE, BABS, FAITH, IMBUDO, KROVANH 6 4 AMBER, YANNI, BILIS Liss), LEKIMA, TORAJI, TRAMI, MORAKOT, MINDULLE, NOCK_TEN, HAITANG, LONGWANG ), TALIM 12 5 HERB, WINNIE, ZEB, KAI_TAK, PRAPIROON, YAGI, RAMMASUN , KUJIRA, MAEMI, MEARI, MEGI, RANIIM, SONGDA, TOKAGE, KHANUN 15 6 GLORIA, SALLY, DAN, MAGGIE, SAM, CHEBI, UTOR, DUJUAN, MELOR (Miller), KOMPASU, OTTO 11 201118764 AERE, DALE, VIOLET, ZANE, ISA, IVAN (Avon), LEVI, OPAL, PETER, ROSIE, TINA, VICK, BART, GLORIA , KATE, OLGA, DAMREY, KIROGI, XANGSANE, CIMARON, HAIYAN, ETAU, LINFA (Lotus), LUPIT (Lu Bi), SOUDELOR (Sudile), CHABA (Jia Ba), MA-ON (Saddle), NIDA (Ni妲), SUDAL (Suda), NABI (Nabi), NESAT ( Nisa), SONCA (Sangka)
由表1發現第7種路徑最多’共有32個;其次為第5 種路徑有15個,而第1、2、3種路徑最少,路徑分類如 下所述: 路徑1.由台灣西南方之南中國海生成之颱風由巴士 海峽約朝東北向進入台灣東南海域並持續遠離台灣者,均 不至對化連港港池造成共振現象。 路徑2:由台灣東北部外海生成之職朝西南向接近 台灣而未直接侵襲花蓮者,亦不至對花蓮港港池造成共振 現象。 路徑 宋島東方海域生成之颱風朝西向 通過呂宋島者’亦不至對花 化連港港池造成共振現象。 路徑4.在菲律賓呂宋氣 6桩w艰 衣馬東方海域生成之颱風朝西北 ^ 連者,影響最為嚴重。 徑5,在菲律賓呂宋鳥 向接近台灣再轉北北西或 ^海域生成之㈣朝西北 甚顯著。 持北由化蓮外海通過者,影響亦 路徑6:在菲律賓呂宋 束方海域生成之颱風朝西北 201118764 向接近台灣由 顯。 巧南端或巴士海峽通過者, 影響亦稱明 •社非律賓呂突I由 北’在距台灣尚遠時 海:生成之趟風朝西 前即轉北北西或轉北由台^過=者’在未接近台灣 東、東北而遠離台灣東部 /通過’ *至轉北北 響。 ° /者,亦稍有影響或甚至無影From Table 1, it is found that the seventh path has a maximum of '32 in total; the second is that there are 15 in the fifth path, and the first, second, and third paths are the least, and the path classification is as follows: Path 1. South of southwestern Taiwan The typhoon generated by the Chinese Sea is caused by the bus strait going northeastward into the southeastern Taiwan Sea and continuing away from Taiwan, and it will not cause resonance to the Hualian Port. Path 2: The position generated by the offshore of the northeastern Taiwan is approaching Taiwan in the southwest and has not directly invaded Hualien. It does not cause resonance in the Hualien Port. Path The typhoon generated by the eastern waters of Songdo Island is heading westward. The people passing through Luzon Island do not cause resonance in Huahua Liangang Port. Path 4. In the Luzon gas of the Philippines, 6 piles of hardships, the typhoon generated by the eastern waters of the Yima River to the northwest ^ Lian, the most serious impact. Trail 5, in the Philippines, the Luzon bird is closer to Taiwan and then turns north-northwest or the sea area (4) to the northwest is very significant. Holding the North by Hualian, the influence of the sea, the path is also 6: The typhoon generated in the Shufang sea area in Luzon, Philippines, is facing the northwest 201118764. The southern end of the road or the passage of the bus strait, the impact is also known as the Ming • social non-legal Lu Du I I from the North in the distance from Taiwan when the sea: the hurricane generated before the turn to the north or north or north by the Taiwan ^ over = Those who are not close to Taiwan's east and northeast and away from eastern Taiwan / pass '* to turn north and north. ° /, also slightly affected or even shadowless
其中,路徑i至路徑3對花蓮港幾乎無 的颱風波浪對花蓮港妒燮 ^ 届斜甚大。上述分類方式係港務 二表浪:翻舶管理而自行制訂,由於本分類法是以 :=:=分_風路徑,經測試較中央氣象局之分 n。a做為模式輸人參數,因此本實施例採用該 知類方式進行模式建構。 颱風風速之分類示如表2。本實施例依照Among them, the typhoon wave from the path i to the path 3 to Hualien Port is very large to the Hualien Port. The above classification method is the Hong Kong service. The two types of waves: the management of the ship is self-developed. Since this classification is based on the :=:= minute_wind path, it is tested by the Central Meteorological Bureau. a is used as a mode input parameter, so this embodiment uses this knowledgeable mode for mode construction. The classification of typhoon wind speed is shown in Table 2. This embodiment is in accordance with
Safflr-Simpson之分類標準,將颱風規模依據颱風中心氣 壓及最大風速分為五級,每級各約有9〜18個。 表2 Saffir:Simpson的跑風規模分類標準 跑風規模 壓力(mb ) 風速(knts ) 1級趙風 >980 64 〜82 2級颱風 965〜980 83 〜95 3級颱風 945〜965 96〜112 4級趟風 920〜945 113〜134 5級趟風 <920 >134 由表3蒲福風級表(Beaufort Scale)可知當風速越大, 波高則越大,風速與波高有著明顯的關係,因此,本實施 201118764 例亦將風速納入影響因子中According to the classification standard of Safflr-Simpson, the typhoon scale is divided into five levels according to the typhoon center air pressure and the maximum wind speed, each of which has about 9 to 18. Table 2 Saffir: Simpson's running wind scale classification standard running wind scale pressure (mb) wind speed (knts) level 1 Zhao Feng > 980 64 ~ 82 level 2 typhoon 965~980 83 ~ 95 level 3 typhoon 945~965 96~112 Level 4 Hurricane 920~945 113~134 Category 5 Hurricane <920 > 134 According to Table 3 Beaufort Scale, the higher the wind speed, the higher the wave height, and the wind speed has a significant relationship with the wave height. In this implementation, 201118764, the wind speed is also included in the impact factor.
用 > 考第1圖,第!圖顯示颱風中心位置距離觀測站 的距離與觀測站的示性波高(Significant Wave Height,Hs) =關係圖’由第i圖可以發現風距離推算點在i儀 上時波高與颱風無明顯關係,若在1500公里以内 有著明顯的關係。本實施例依花蓮港之船舶 動態分別為: 术期間’共分成四種船隻 ⑴花蓮料之船隻與其裝卸作f皆未受影響。 201118764 (2) 花蓮港内產生湧浪,但湧浪並未大到影響裝卸作 業,船隻亦不需出港避風。 (3) 產生之湧浪大到船隻需出港避風,才能免除斷纜 的可能。 (4) 船隻產生了斷纜。 船隻發生斷纜可能造成船隻在港内任意漂泊進而造 成船隻更嚴重之船隻異動,如可能與港内其他之船隻產生 破撞等。因此,我們將斷纜歸於最嚴重之船隻異動指數4, 並依據對船隻異動之嚴重性,將定義為四個等級的船隻異 動指數(Index For Ship Escape,ISE) ° 根據此4個船隻異動指數,並將82個颱風加以分類, 示如表4,由表4可知82場颱風中有27場造成船隻的斷 纜,有25場颱風對船隻沒有造成影響,12場颱風有湧浪 的產生,1 8場颱風船隻需出港避風的情形。 表4颱風之船隻異動指數歸類表 船隻異動 指數ISE 颱風名稱 個數 1 BART(巴特)、BETH(貝絲)、BOPHA(寶 發)、CAM(凱姆)、CIMARON(西馬隆)、 CONSON(康森)、DAMREY(丹瑞)、 ERNIE(爾尼)、FAITH(費絲)、GLORIA(葛 樂禮)、HAITANG(海棠)、HAIYAN(海 燕)、KATE(|iL 4寺)、KOMPASU(康柏 Λί )、 KROVANH(柯羅旺)、LEVI(里維)、 MEGI(梅姬)、MELOR(米勒)、 NANGKA(南卡)、NAR1(納莉)、 NOGURI(諾古力)、PRAPIROON(巴比 侖)、RANANIM(蘭寧)、TRAMI(潭美)、 VICKI(維琪) 25 2 LEKIMA(利奇馬)、L1NFA(蓮花)、 MA_ON(馬鞍)、MAEMI(梅米)、 MEARI(米雷)、MORAKOT(莫拉克)、 12 201118764 NESAT(尼莎)、OLGA(歐佳)、PETER(彼 得)、SONCA(桑卡)、TINA(蒂納)' YANNI(楊妮) 3 CHABA(佳芭)、CHEBI(奇比)、DALE(戴 兒)、DAN(丹恩)、DUJUAN(杜鵑)、 KAI_TAK(啟德)、KHANUN(卡努)、 KIr5gi(4 洛基)、KUJIRA(柯吉拉)、 MAGGIE(瑪姬)、NIDA(妮妲)、 ]^0(:反_丁£1^(納坦)、110 51£(羅西)、 SAM(山姆)、SUDAL(舒達)、UTOR(尤 特)、VIOLET(魏萊特)、YAGI(雅吉) 18 4 AERE(艾利)、AMBER(安 ί白)、BABS(芭 比絲)、BILIS(碧利斯)、ETAU(艾陶)、 GLORIA(葛樂禮)、HERB(賀伯)、 IMBUDO(尹布都)、ISA(麗莎)、IVAN(艾 文)、1^€^0从八\0(龍王)、1^卩1丁(盧碧)、 MINDULLE(敏督利)、NABI(娜比)、 OPAL(歐珀)、OTTO(奥托)、 RAMMASUN(雷馬遜)、SALLY(莎莉)、 50、00八(桑達)、50110£[011(蘇迪勒)、 TALIM(泰利)、TOKAGE(陶卡基)、 TORAJI(桃芝)、WINNIE(溫妮)、 XANGSANE(象神)、ZANE(薩恩)、 ZEB(瑞伯) 27Use > test the first picture, the first! The figure shows the distance from the observation center of the typhoon center to the observation station. (Significant Wave Height, Hs) = relationship diagram. From the i-th diagram, it can be found that the wind distance is not significantly related to the typhoon when it is on the i-meter. If there is a clear relationship within 1500 km. The dynamics of the ship in Hualian Port in this embodiment are as follows: During the operation period, the ship is divided into four types of vessels. (1) The boat of Hualien material and its loading and unloading work are unaffected. 201118764 (2) There are swells in the port of Hualien, but the swells are not so large that they affect loading and unloading operations, and vessels do not need to leave the port to avoid the wind. (3) The swell generated is so large that the ship only needs to leave the port to avoid the wind, so as to avoid the possibility of cable breakage. (4) The vessel has broken the cable. A cable breakage in a vessel may cause the vessel to drift freely in the port and cause a more serious vessel change, such as a possible collision with other vessels in the port. Therefore, we attribute the cable to the most severe vessel transaction index 4, and based on the severity of the vessel movement, it will be defined as four levels of Index For Ship Escape (ISE) ° based on the four vessel transaction index 82 typhoons are classified as shown in Table 4. It can be seen from Table 4 that 27 of the 82 typhoons caused the ship to be broken, 25 typhoons did not affect the ship, and 12 typhoons caused swells. The 18 typhoon boats only need to be out of the harbor to avoid the wind. Table 4 typhoon vessel transaction index classification table vessel transaction index ISE typhoon name number 1 BART (Bart), BETH (Beth), BOPHA (Baofa), CAM (Kem), CIMARON (Ximalong), CONSON (Conson), DAMREY, ERNIE, FAITH, GLORIA, HAITANG, HAIYAN, KATE (|iL 4 Temple), KOMPASU (Compaq) Λί ), KROVANH, LEVI, MEGI, MELOR, NANGKA, NAR1, NOGURI, PRAPIROON Billenn, RANIIM, TRAMI, VICKI 25 2 LEKIMA, L1NFA, MA_ON, MAEMI, MEARI ), MORAKOT, 12 201118764 NESAT, OLGA, PETER, SONCA, TINA YANNI 3 CHABA , CHEBI, DALE, DAN, DUJUAN, KAI_TAK, KHANUN, KIr5gi (4 Rocky), KUJIRA, KUJIRAMAGGIE, NIDA, ^0(: 反_丁£1^(纳坦), 110 51£(罗西), SAM(萨姆), SUDAL(舒达), UTOR(尤尤Special), VIOLET, YAGI 18 4 AERE, AMBER, BABS, BILIS, ETAU, GLORIA ), HERB (He Bo), IMBUDO (Yim Budu), ISA (Lisa), IVAN (Ivan), 1^€^0 from eight \0 (Dragon King), 1^卩1 Ding (Lu Bi), MINDULLE, NABI, OPAL, OTTO, RAMMASUN, SALLY, 50, 00 (Sanda), 50110 £ [ 011 (Sudile), TALIM, TOKAGE, TORAJI, WINNIE, XANGSANE, ZANE, ZEB 27
故於本實施例中,將船隻異動指數1〜4定義如下: 船隻異動指數1 :花蓮港内之船隻與其裝卸作業皆未 受影響。 船隻異動指數2 :花蓮港内產生湧浪,但湧浪並未大 • 到影響裝卸作業,船隻亦不需出港避風。 船隻異動指數3 :產生之湧浪大到船隻需出港避風, 才能免除斷纜的可能。 船隻異動指數4 :船隻產生了斷纜。 接著,對於颱風風速及颱風中心至花蓮港之方位角對 船隻異動指數進行分析,請同時參考第2-1圖與第2-2圖。 第2-1圖顯示颱風風速與船隻異動指數關係圖,第2-2圖 顯示颱風中心至測站方位角與船隻異動指數關係圖,由第 201118764 2-1圖可看出船隻異動指數1分布於75〜175knots的範圍, 指數2〜4則其風速分布在55〜295 knots的範圍,指數2〜4 之颱風風速指數1之風速較無鑑別度,顯示颱風風速必非 直接影響船隻異動。另外’由第2-2圖可知,影響花蓮港 船舶動態,以花蓮港之緯度為界〇〜150度及-60〜0度,且 各指數之分佈範圍約略相似。由風速及方位角對異動指數 之關係,可知兩者之分佈較分散,雖無一定之規則及不易 辨識’但該兩參數仍是可定性描述影響異動指數之關係。Therefore, in the present embodiment, the vessel transaction index 1 to 4 is defined as follows: Vessel transaction index 1: The vessel in Hualien Port and its loading and unloading operations are unaffected. Vessel Change Index 2: There is a swell in the Hualien Port, but the swell is not large enough to affect the loading and unloading operations, and the vessel does not need to leave the port to avoid the wind. Vessel Change Index 3: The swell generated is so large that the ship only needs to leave the port to avoid the wind, so as to avoid the possibility of cable breakage. Vessel Change Index 4: The vessel has broken the cable. Next, for the typhoon wind speed and the azimuth angle of the typhoon center to Hualien Port, analyze the vessel movement index. Please refer to Figures 2-1 and 2-2. Figure 2-1 shows the relationship between the typhoon wind speed and the ship's transaction index, and Figure 2-2 shows the relationship between the typhoon center-to-station azimuth and the ship's transaction index. From the 201118764 2-1 chart, the ship's transaction index 1 distribution can be seen. In the range of 75~175knots, the index 2~4 has a wind speed distribution in the range of 55~295 knots, and the wind speed of the typhoon wind speed index 1 of index 2~4 is less discriminating, indicating that the typhoon wind speed must not directly affect the vessel movement. In addition, it can be seen from Figure 2-2 that the dynamics of Hualien Port are affected by the latitude of Hualien Port, which is ~150 degrees and -60~0 degrees, and the distribution range of each index is similar. From the relationship between wind speed and azimuth to the transaction index, it can be seen that the distribution of the two is relatively scattered, although there are no rules and it is difficult to identify 'but the two parameters are still qualitatively describe the relationship affecting the transaction index.
除此之外,選用颱風與花蓮港距離(£))、颱風最大風速 (匕-X)、颱風中心至花蓮港之角度(0 〇、颱風行進方位角 2)、風場能量五n(Fma,/Z叹(D))與中央氣象局發佈之海、陸 上警報(F/, Waring Index)等六個影響船隻動態的颱風因子 建立颱風與ISE間的關係,其相關因子之示意圖如第2d 圖所不’第2-3圖顯示影響船舶動態之颱風因子示意圖。 其中,海上颱風警報定義為:颱風七級暴風圈二十四小時 2進入臺灣或金門'馬袓海岸線一百公里以内海域時發 △,之後每三小時發報一次,必要時得加發之。海、陸上 趟風警報定義為:關㈣域暴風圈十人小時後進入臺 灣或金門、馬祖陸上時發布,之後每一小時發報一次。必 要時^同,發布海上陸上趟風㈣,不受上列條件所規 臺&灣或^風警報定義為:當趙風的七級風暴風範圍離開 :金門、馬袓陸地’但仍未離一百公里近海時,改發 解除::警:ή當確定趟風離開—百公里近海時,即發布 解轉向或消散時,得直接解_風警報。 201118764 在本實施例令,所應用類神經網路具有學習最佳化的 功能’並透過其模式輸出值與學習目#值間的連結建立起 其相關性,利用網路權重(weights)與門限值(bias)來表 關係的強弱。 、本發明-實施例中’所選取之網路演算法則為倒傳遞 法(Back-propagation),由於倒傳遞網路具有監興 習,神經元間的交互作用經由該學習方式求得模式輸出值 與實際值之最小均方根誤差。本文架構神經網路使用 ♦ MaUab軟體來建置倒傳遞類神經網路,如第3圖所示,第 3圖顯示類神經網路架構示意圖。 倒傳遞學習演算法是被廣泛使用的一種學習演算 法,因其具有學習及回想的功能,故可在回想過程中以相 同於學習過程的方式來進行預測。一般倒傳遞網路可分為 三部份,輸入層(Input Layer)用以接受外在環境的訊息, 隱臧層㈤編一叫則表現輸入與輸出層(Output Layer) 各處理單元間的相互關係,並以權重和閥值來聞述該相關 •性:輸出層用以輸出訊息給外在環境。倒傳遞網路學習方 式是由輸入相當數量之學習樣本,應用向前饋入與誤差向 後修正兩步騾,推求輸入變數與輸出變數的内在對映規 則,再應用回想功能,進行新案例之輸出變數值推估。 第3圖架構之網路為一層隱藏層及一層輸出層,巧隨 為轉換函數,s h為隱藏層及輸出層之神經元個數,立 扮演推論結果經由轉換函數映射之過程為輸入變數矩 陣’ W口卜為輸入層與隱藏層間的權重和闕值矩陣,^ 12 201118764 和卜為輸出層與隱藏層間的權重和閥值矩陣。經由網路學 習誤差最佳化的過程,神經網路將輸入值及學習目標值的 關係紀錄在間值與權重上,以此表示輸入值與學習目標值 r::的關係強弱。其中,類神經網路使用轉換函數使隱藏層 模仿生物神經元處理非線性的機能,以輸入值之加權乘積 成處理單7"之輸出值。整個倒傳遞網路模式由隱藏 層與輸入參數及輸出層的關係可用方程式表示為 鲁 Λι,*χΐ - Λχ, {lWtSXn Pnxl + bUsx]) ⑴In addition, the distance between Typhoon and Hualien Port (£), the maximum wind speed of typhoon (匕-X), the angle from Typhoon Center to Hualien Port (0 〇, typhoon azimuth 2), wind field energy n n (Fma , /Z sigh (D)) and the typhoon factor affecting vessel dynamics, such as the sea and land warning issued by the Central Meteorological Administration (F/, Waring Index), establishes the relationship between typhoon and ISE, and the related factors are shown in Figure 2d. Figure 2-3 shows a schematic diagram of the typhoon factor affecting ship dynamics. Among them, the typhoon warning at sea is defined as: Typhoon 7-level storm circle 24 hours 2 When entering Taiwan or the Golden Gate 'Malay coastline within 100 kilometers of the sea area, △, then every 3 hours to report, if necessary, add it. The sea and land hurricane alarm is defined as: off (4) when the storm circle enters Taiwan, Jinmen, and Mazu land after ten hours, and then reports every hour. When necessary, the same as the release of the land hurricane (4), not subject to the above conditions, the regulation of the Bay & ^ wind alarm is defined as: when Zhao Feng's seven-level storm winds leave: Kinmen, Maji land 'but still not When it is a hundred kilometers away from the sea, the change is lifted:: Police: When the hurricane is determined to leave - 100 kilometers offshore, when the release of the turn or dissipate, you must directly solve the _ wind alarm. 201118764 In this embodiment, the applied neural network has the function of learning optimization' and establishes its correlation through the connection between its mode output value and the learning target value, using network weights and thresholds. The value of the bias is the strength of the relationship. In the present invention, the selected network algorithm is a back-propagation method. Since the reverse transmission network has a monitoring function, the interaction between the neurons obtains the mode output value through the learning method. The minimum root mean square error from the actual value. In this paper, the neural network uses ♦ MaUab software to build a reverse-transfer-like neural network. As shown in Figure 3, Figure 3 shows a schematic diagram of a neural network architecture. The inverse transfer learning algorithm is a widely used learning algorithm. Because it has the function of learning and recalling, it can be predicted in the same way as the learning process in the process of recalling. The general inverted network can be divided into three parts. The input layer is used to accept the information of the external environment. The hidden layer (5) is called to represent the mutual input and output layers. Relationships, and the correlations are described by weights and thresholds: the output layer is used to output messages to the external environment. The inverse transfer network learning method is to input a considerable number of learning samples, apply forward feed and error backward correction two steps, deduct the internal mapping rules of input variables and output variables, and then apply the recall function to output the new case. Variable numerical estimation. The network of the architecture of Figure 3 is a layer of hidden layer and an output layer. It is a conversion function, sh is the number of neurons in the hidden layer and the output layer, and plays the role of inference. The process of mapping through the conversion function is the input variable matrix. W is the weight and 阙 matrix between the input layer and the hidden layer, ^ 12 201118764 and Bu is the weight and threshold matrix between the output layer and the hidden layer. Through the process of network learning error optimization, the neural network records the relationship between the input value and the learning target value on the inter-value and weight, thereby indicating the relationship between the input value and the learning target value r::. Among them, the neural network uses a conversion function to make the hidden layer mimic the function of the biological neuron to deal with nonlinearity, and the weighted product of the input values is the output value of the processing single 7". The relationship between the hidden layer and the input parameters and the output layer of the entire inverted network mode can be expressed by the equation as Λ Λ,, χΐ - Λχ, {lWtSXn Pnxl + bUsx]) (1)
a2,r,l ~/rxl(^,rXiei ixI :中…為隱藏層之輸出值矩陣’幻為倒傳遞網路輸出 值矩陣,/為轉換函數。藉由 稚田工八之運算後求出網路輸出 值矩陣,其與學習目標矩陣^ 邊方定義兩矩陣 相減之長度(N_)為誤差函數£,其表示 J〜-y«2 N (3)A2,r,l ~/rxl(^,rXiei ixI :in...is the output matrix of the hidden layer' illusion as the inverse output network output value matrix, / is the conversion function. It is obtained by the operation of the child The network output value matrix, which is defined by the learning target matrix ^ edge defined by the two matrices (N_) is the error function £, which represents J~-y«2 N (3)
因此’整個網路學習的過转A • ^ . 于㈢的過私為了最佳化各神經元間之 權重與閥值使誤差函數達到最小值 政#山“ 思j取j值’反覆迭代其過程使網 路輸出值趨近學習目標值。含網 ^ 1 '為路疋成學習過程(最佳化過 程)’建立最合適的權重和閥值後即 交卩了以此網路的回想過程 ^進仃預心析。以下說明其演算過程,最佳化過程可分 為兩類’-為啟發式的最佳化’如可變學習數率(― ^annng Rate)與㈣性(Resilient)之演算* ;另—為使用 “準的數值最佳化,如共軛梯 .u ,、挑梯度法、擬牛頓法及 ——Μ,法。—般而言,在函數逼近的 13 201118764 問題上Levenberg-Marquardt演算法將有最快之收斂性 (Neural Network Toolbox User’s Guide)。因此,本文選擇 使用L-M演算法做為達到網路誤差函數最小之最佳化過 程’該演算法使用Hessian矩陣逼近方式來達到最佳的目 的,並以下式表示之 ^k+\ =^k ~\JTJ + ^]~lJTe (4)Therefore, 'the whole network learning is overturned A ^ ^ . (3) The over-privacy in order to optimize the weights and thresholds between the neurons to make the error function reach the minimum value of the political value of the mountain The process brings the network output value closer to the learning target value. The network ^ 1 'to establish the most appropriate weights and thresholds for the learning process (optimization process) is the process of recalling the network. ^Introduction to the heart. The following describes the calculation process, the optimization process can be divided into two types - "for heuristic optimization" such as variable learning rate ("annng rate" and (four) (Resilient) The calculation *; another - to use the "quasi-value optimization, such as conjugate ladder .u, pick gradient method, quasi-Newton method and - Μ, method. In general, the Levenberg-Marquardt algorithm will have the fastest network convergence (Neural Network Toolbox User’s Guide) on the 13 201118764 problem of function approximation. Therefore, this paper chooses to use LM algorithm as the optimization process to minimize the network error function. The algorithm uses Hessian matrix approximation to achieve the best purpose, and the following formula represents ^k+\ =^k ~\ JTJ + ^]~lJTe (4)
其中,沿、幻Η·/分別表示每次疊代權重{心,^^與 門限值{h,~}之前後最佳化計算值,y為Jac〇bian矩陣, 其包含網路誤差與權重值和閥值的一階微分,e為網路誤 差向量,/為單位向量。式(4)之〆=〇時,則近似牛頓法的 Hessian矩陣,當〆值很大時,式(4)則具有較小步階之梯 度下降’其目的即為盡可能之快速地移向牛頓法以達到網 路誤差最佳化的結果,最後可獲得適應該網路之最佳化權 重和閥值。因此,倒傳遞類神經網路經由式(丨)及式(?)叶 算輪入值經轉換函數映射制模式輸出值,並以式(4)作為權 重與門限值之最適化反覆迭代計算,其網路學習完成之標 準則以式(3)來判斷。 ^ 由分析船隻動態資料得知,熱帶性低氣壓均未對花蓮 港船隻造成影響,因此’所述之颱風影響船舶動態參數, ,風與花蓮港距離⑼、趟風最大風速(U、風中心至 化蓮港之角度(61丨)、颱風行進方位角(Θ 2)、風場能量 與中央氣象局發佈之海、 :個影響船隻動態的跑風因子及中央氣象局㈣報= 未來48小時的預測資料,做為類神經網路之輪入值。 201118764 本模式將以氣象局或JTWC發佈 作為類神經模式之相關輸入值,二關參數結果 於固定區域_風與船隻異動指c構,對 風資料Z)、r 、 e 可以表不為颱 'θΐ、θ2ϋ函數的時序列組合 式(3-5)中%為係s /丨為函數。·為時間。Wherein, the edge, the illusion·/ respectively represent the optimization value of each iteration weight {heart, ^^ and the threshold value {h,~}, and y is the Jac〇bian matrix, which contains the network error and weight. The first-order differential of the value and the threshold, e is the network error vector, and / is the unit vector. When 〆=〇 of equation (4), it approximates the Hessian matrix of Newton's method. When the value of 〆 is large, equation (4) has a gradient of smaller steps. The purpose is to move as fast as possible. Newton's method is used to optimize the network error, and finally the optimal weights and thresholds for the network can be obtained. Therefore, the inverse transfer-like neural network calculates the mode output value through the conversion function mapping mode via the formula (丨) and the formula (?), and uses equation (4) as the optimization of the weight and the threshold value. The standard for completing the network learning is judged by the formula (3). ^ According to the analysis of the dynamic data of the vessel, the tropical depression has not affected the Hualien port vessel, so the typhoon described affects the dynamic parameters of the vessel, the distance between the wind and Hualien port (9), and the maximum wind speed of the hurricane (U, wind center The angle of Yanghua Port (61丨), the azimuth of typhoon (Θ 2), the energy of the wind field and the sea issued by the Central Meteorological Administration, the running wind factor affecting the dynamics of the ship and the Central Meteorological Bureau (4) report = the next 48 hours The forecast data is used as the round-in value of the neural network. 201118764 This model will be published by the Meteorological Bureau or JTWC as the relevant input value of the neurological model, and the second parameter will be the fixed area _ wind and ship movements The wind data Z), r, e can be expressed as a time series combination (3-5) of the 'θΐ, θ2ϋ function as a function of the system s / 丨. · For time.
本實施例選擇之學習範例為含蓋4魏風種類、4種 ::又異動指數’選取邀風共5(Μ固,做為類神經之學習範 :卜本實施例採用2層隱藏層來表現非線性的效果,網路 架構為輸入神經元為6個,輸出層為!個為值,由經 :與嘗試之測試結果’本實施例之2層隱藏層,其神經元 =為80及40個。隱藏層轉換函數則採用對數雙彎曲轉換 函數(Log sigmoid transfer functi〇n),而最終迭代次數設定 為1500代,在誤差平方近似於〇或迭代次數為Η⑼代時 之任何-個條件’該網路即停止學習,但本發明不應以此 為限。 本實施例之圖形使用者介面(Graphical User Interface, GUI)環境為 Matlab 中的 GUIDE(Graphicai User InterfacesThe learning example selected in this embodiment is the cover type 4 Wei Feng type, 4 kinds:: and the change index 'to select the wind to be a total of 5 (sturdy, as a nerve-like learning model: the present embodiment uses two hidden layers to The effect of non-linearity is as follows: the network architecture is 6 input neurons, and the output layer is !, the value is: and the test results of the test are 'the hidden layer of the second layer of this embodiment, the neuron=80 and 40. The hidden layer transfer function uses Log sigmoid transfer functi〇n, and the final iteration number is set to 1500 generations, any condition where the error square is approximated to 〇 or the number of iterations is Η(9) generation. 'The network stops learning, but the invention should not be limited to this. The graphical user interface (GUI) environment of this embodiment is GUIDE in Matlab (Graphicai User Interfaces)
Devel〇pment Environment)。操作環境為 Mathw〇rks 公司的 科技運算應用軟體Matlab,版本支援Matlab 5 〇以上系統。 圖形化介面的開發目標期望能以最少的操作程序作 最多的流程以及展現最多的資訊,如此一來可以減少視窗 的切換以及提升操作流程的流暢度。 201118764 1例中,整個圖形化介面主要由五個視窗組成、分 紀風路^介面視窗2'資料輸人視窗3.資料存檔視窗4 免風路徑選擇視窗 4· 圖,八E,人 ·即時實測“圖6.即時衛星影 圖,分別介紹如下: ^像 1.主介面視窗 在視(二見窗標題列:標題列明確列出目前介面所 在視岛並§主明介面版本等資訊。 吓 ⑺介面主視窗工具列:工具列包含資料輸入輸出、 Γ炉訊以及輔助四個部分。工具列的編排主要符 合一身又Windows應用藉&加+致 .丄 浐式的架構,以樹狀的結構將所要選 取的扣令存放於内,如第 、 主視窗。 如第4-1圖所不’第4-丨圖顯示介面 :同時參考第4-2圖,,4_2圖顯示樹狀結構的工且 列’第:子工具列為輸入與輸出部分包括颱風資料的輸 入、。又異動指數的輸出及離開本介面。第二子工具 模擬部分,此部份為資料給 ’ 枓輸入確疋無誤後選取適當的類神 經架構進行模擬,模擬後可 、聚便了 k出坪細的船隻異動指數報 告。第三子工具列為趟風資訊部分’透過網際網路取得即 時的颱風資訊。第四子工呈 卞具列為說明以及版權宣告,包含 呼叫完整的使用手冊以及關 及關於本介面的版本資訊視窗開 啟’使用手冊的開啟將今叫外部程式以開啟HTML資料的 瀏覽器。 ⑴輸入資料之趟風距離變化圖:此區域可即時將所 輸入的跑風位置資料運算錢風與港口之間距離變化並 16 201118764 以繪圖方式來表示, 料以及趟風最接近點 幫助使用者即時了解所輸入之颱風資 的時間。 ⑷冑人資料之顧風速變化圖:此區域可即時 輸广風風速資料以綠圖方式來表示,幫助使用者即時 貝科並軚不出風速最大值所發生的時 間0 (5)颱風路徑圖:颱風路徑圖視窗中展示該颱風在北緯 15 30度、東經u 4〜138度範圍内的賤風行進路徑。 ()也隻異動&數.此為本模式讀取跪風資訊後進行模 擬而推算之船隻異料n在模式推估結果到達3級的 刖A小時發布警報音效’圖中紅色垂直線顯示警報發布時 間,綠色垂直線則表示現在的時間。本實施例以現有資料 進行船隻異動指數的推算後仍以15#、—次地即時比對現 在的時間,當綠色垂直線與紅色垂直線重合時即發布警報 音效。 (7)警示燈:隨著船隻異動指數不同,利用閃光三色燈 參來發布目前狀況,當無燈號時代表船隻異動指數為i,即 花蓮港内之船隻與其裝卸作業皆未受影響;綠燈表示有船 隻異動指數為2,花蓮港内產生湧浪,但湧浪並未大到影 響裝卸作業,船隻亦不需出港避風;黃燈代表船隻異動指 數3,港内產生之湧浪大到船隻需出港避風,才能免除斷 纜的可能;紅燈表示船隻異動指數4,依據模式訓練時所 用的歷史資料顯示船隻產生了斷纜,此時代表花蓮港内的 停泊船隻相當危險。此外,在警示燈右側並提供使用者調 17 201118764 整警報音效之音量。 (8)警不標誌:利用飛入的方法提醒使用者目前船隻 異動指數的等級以及船舶動態之建議,共四種警示標誌, 如第4-3圖所示,第4_3圖顯示船舶異動指數所對應之警 示標誌。在一實施例中,各個警示標誌中的右側註明了船 隻異動指數以Level 1〜4表示,左側的圖示則大略表達目 前港内船舶狀況,LeVel Ϊ為港内海水面靜穩;Level 2為 有湧浪但仍不影響船隻裝卸作業;Level 3則表示港内產生 之渴浪大到船隻需出港避風,才能免除斷纜的可能;Levei 4則表示船隻可能已產生了斷纜,十分危險。 2.資料輸入視窗 如第5·1圖所示,資料輸入視窗與一般Wind〇ws檔案 開啟視窗相似,操作部分詳述於下: (1) 目前資料夾位置:顯示目前所在資料夾,使用者 能在此區域切換各個位於該電腦中的儲存設備或是網路 儲存設備以更改目前資料夾位置。 (2) 回上一頁按鈕:本按鈕功能為回到之前所在目 錄0 (3) 回上一層按鈕:本按鈕功能為回到目前所在資料 夾的上-層’若是以處在最上層資料夾則會回到磁碟根目 錄,Windows作業系統根目錄的上一層則為我的電腦。 (4) 新增資料夾按鈕:本按鈕能在所在資料夾内再新 增一子資料夾。 (5) 檢視模式變換按鈕:本按叙可將目前所在資料失 201118764 内的檔,表現方式做適當的改變,包含大型圖#、小型圖 不、清單、詳細資料以及縮圖五種方式。 (6)㉟案名稱區域:使用者可在本區域輸入欲開啟的 輸入資料檔名。 —⑺it取檔案類型:本區域能夠設定檔案列表區内顯 :的槽案類型’本介面在此内定為顯示附檔名為*遍之所 有檔案,使用者可視需要更改為顯示所有擋案。 ⑻Μ啟檔案按紐:本按钮可在選取檔案後執行開啟 的動作’並回到主視窗進行下―步的運算以及檢核。 (9)取4按紐·不作任何更動離開本視窗。 _肖案區列表區:列出本資料夾中所㈣案以及子 1)說明按鈕’按下此按鈕後滑鼠游標會主現問號 再點及本視窗中各個元件後’會出現各元件的簡單說明 (12)關閉視窗按鈕:同於取消按鈕,不作任何改變 開本視窗。 3.資料存檔視窗: 第5 2圖所不,第5_2圖顯示模擬結果資料存檔 動少叙’、歹^輸入與輸出’,中的”檔案儲存,,可進行船隻 、 仔棕視®,視窗中各元件介紹同: '料輸入視窗。而存槽的#宏 扪棕案格式限制以及更動都是在: 圖中(7)存槽類型的區域 ,Λ ^ 埤兩個不同的視窗會對存檔t 格式作限制,使用者盔須轳 …貝擔〜會將ascii的資料檔儲存; 雜誤的格式。 19 201118764 4.颱風路徑選擇視窗 ^當資料輸入完成之後,可選取上方工具列之,,模式運 算中的進订推算”來進行模式推算,此實模式會自動列出 7種路徑提供使用者做選擇。請參考第5·3圖第5_3圖 ”’、颱風路各選擇視窗’此7種分類是由表2中花蓮港務 局所訂定。 以下舉例説明該模式完整操作流程: 在Matlab環境令可利用工具列的槽案開啟功能或是 接於C_and WindGW直接輸人则介面檔名開啟本介 ’擋名目前暫定為mship.p(如第6]圖所示)。 進入㈣動態預警模式GUI後,可看到主工作區有四 個圖塊,以及警示燈號,但由 ...s 禾輸入資料進行模擬推算 無法顯示結果。在上方工具列中此 姐,,& 八幻中目刖僅能選擇,,讀入資 枓進行·貝料輸入,或選擇”外部 衞足旦m 矾中的即時波高與即時 衛星影像來查詢資料’以及 宫土,,.而甘从 及"他中的”使用手冊,,與,,版權 H其具列選項如”資料儲存,,以及,,模式運算” 這是為了避免使用者錯誤的摔作=反白狀態無法選擇, 報告。在此步驟我們可以選”輸入與 :果 進行颱風資料的讀取。並依昭第 ,w入貝料’, 入視窗選取輸入資料。此外,:與第6·3圖中的輸 該主機硬碟中或是網路硬碟中的中可切換 名内定為(*.dat),使用者可調整幸#备案之副檔 正棺―型(”)來尋找欲輸 20 201118764 入的任何ascii檔案名稱。目前颱風資料輸入格式採用最 簡單的純文字文件,文件中只要包含以下列颱風資料:I 寺門序歹丨2.年3.月4.曰5.時6.經度7.緯度8·風速9·趙風警 報狀態,格式並無嚴格限制,使用者可利用EXCEL或是 文書編輯軟體都可以做成此輸入檔,每列資料中可用 Tab、Space、逗點等用來表示間隔。若是資料中有遺漏則 需要由操作者先做内插或補遺處理,處理後調整時間序列 (由〇開始至最後一筆資料)即可開始做輸入。 • t颱風資料輸入完成後若data長度與格式無問題本 "面會自動將雖風中心最大風速、趟風中心與花蓮港距離 以及颱風路徑圖繪圖於本介面之視窗圖塊内,如第6_4 圖而在檢視二張圖沒問題後即可點選工具列的,,模式運 算中的進行推算”進行路徑選擇後開始模擬推算。如第 圖所示為推算完成後的狀況,介面右下方即為船隻 異動指數隨時間的變化圖,而介面又方除了警示燈之外還 有警不標誌。推算完成之後即可利用上方工具列,,輸入與輸 # 出”中的”資料儲存,,進行儲存。 綜上所述,船舶動態預警系統視窗化操作模式不但提 供使用者在操作上的便利,並且整合了整個模式的運作, 整σ後的模式可&升對於未來自動化環境的配合度。未來 可應用於提供網路即時船舶動態模式推算,可以在無人操 作的環境中直接擷取颱風觀測資料以推算出船隻異動指 數並同時以該伺服器作即時發布。Devel〇pment Environment). The operating environment is Mathw〇rks's technology computing application software Matlab, and the version supports Matlab 5 〇 or higher. The development goal of the graphical interface is to achieve the most processes and the most information with the fewest operating procedures, which can reduce the switching of windows and improve the smoothness of the operation process. 201118764 In 1 case, the entire graphical interface consists of five windows, divided into sections, and the interface window 2' data input window 3. Data archive window 4 Free wind path selection window 4 · Figure, eight E, person · instant Measured "Figure 6. Instant satellite imagery, respectively, as follows: ^ Like 1. The main interface window is in view (see the window title column: the title column clearly lists the current interface on the island and § the main interface version and other information. Scared (7) Interface main window toolbar: The toolbar contains four parts: data input and output, Γ furnace and auxiliary. The layout of the toolbar is mainly in line with the Windows application and the structure of the tree structure. Store the deductions to be selected, such as the first and main windows. If the picture is not shown in Figure 4-1, the display interface is shown in Figure 4-2, and the 4_2 figure shows the structure of the tree structure. And the column 'the sub-tools are the input and output parts including the input of the typhoon data, and the output of the transaction index and leaving the interface. The second sub-tool simulation part, this part is the data to the '枓 input is correct Choose the appropriate class god The structure is simulated. After the simulation, the vessel's transaction index report is collected. The third sub-tool is listed as the Hurricane Information section to obtain instant typhoon information through the Internet. For explanation and copyright announcement, including the complete user manual of the call and the version information window about the interface, the opening of the user manual will be called the external program to open the HTML data browser. (1) Change the hurricane distance of the input data : This area can instantly calculate the distance between the wind and the port calculated by the entered wind position data and 16 201118764. The material and the closest point of the hurricane help the user to know the time of the entered typhoon. (4) The wind speed change map of the deaf person data: This area can be used to display the wind and wind speed data in real time. It helps the user to instantaneously and does not know the time when the maximum wind speed occurs. 0 (5) Typhoon path Figure: The typhoon path window shows the hurricane travel path of the typhoon at 15 30 degrees north latitude and 4 to 138 degrees east longitude. Dynamic & number. This is the model for the model to read the hurricane information and then simulate the ship dissimilar material n in the mode estimation result reaches the level 3 刖A hour release alarm sound effect 'the red vertical line in the figure shows the alarm release time, The green vertical line indicates the current time. In this embodiment, the current data is used to calculate the ship's transaction index, and the current time is still compared with the current time of 15#, the second time, and the alarm is issued when the green vertical line coincides with the red vertical line. (7) Warning light: With the different transaction index of the vessel, the flashing tri-color lamp is used to release the current situation. When there is no signal, the vessel's transaction index is i, that is, the vessel in Hualien Port and its loading and unloading operations are unaffected. The green light indicates that there is a vessel transaction index of 2, and there is a swell in the Hualien port, but the swell is not so large that it affects the loading and unloading operations, and the vessel does not need to leave the port to avoid the wind; the yellow light represents the vessel's transaction index 3, and the swell generated in the port is large enough to the vessel. It is necessary to avoid the wind in order to avoid the possibility of cable breakage; the red light indicates the ship's transaction index 4, and the historical data used in the training according to the mode indicates that the ship has broken the cable. At this time, it is quite dangerous to represent the mooring vessels in Hualien Port. In addition, on the right side of the warning light and provide the volume of the user's adjustment of the 2011 18764 full alarm sound. (8) Warning sign: Use the method of flying in to remind the user of the current level of the ship's transaction index and the recommendations of the ship's dynamics. There are four warning signs, as shown in Figure 4-3. Figure 4_3 shows the ship's transaction index. Corresponding warning signs. In one embodiment, the right side of each warning sign indicates that the vessel transaction index is represented by Level 1 to 4. The icon on the left side roughly indicates the current state of the vessel in the port, and LeVel is the static state of the water in the harbor; Level 2 is a surge. Waves still do not affect the loading and unloading operations of the vessel; Level 3 means that the thirsty waves generated in the port are so large that the ship can only avoid the wind and the wind can be removed, and the Levei 4 indicates that the vessel may have broken the cable, which is very dangerous. 2. Data input window As shown in Figure 5.1, the data input window is similar to the general Wind〇ws file open window. The operation details are detailed below: (1) Current folder location: display current folder, user In this area, you can switch between the storage devices or network storage devices located in the computer to change the current folder location. (2) Back to previous page button: This button function is to return to the previous directory 0 (3) Back to the previous layer button: This button function is to return to the upper layer of the current folder - if it is in the top folder It will go back to the root directory of the disk, and the upper layer of the Windows operating system root directory is my computer. (4) Add folder button: This button can add a new sub-folder in the folder where it is located. (5) View mode change button: This can be used to change the current data in 201118764, and the performance mode is appropriately changed, including large map #, small map, list, detailed information and thumbnail. (6) 35 Name field: The user can input the name of the input data file to be opened in this area. —(7)it file type: This area can set the type of the slot in the file list area. The interface is defined here to display all files with the file name *over, and the user can change it to display all files as needed. (8) 档案 档案 File button: This button can perform the action of opening after selecting the file ’ and return to the main window for the next step operation and check. (9) Take the 4 button and leave the window without making any changes. _Shaw case list area: List the (4) case in this folder and sub-1) Description button 'When this button is pressed, the mouse cursor will be the main question mark and then the various components in this window will appear. Brief description (12) Close the window button: Same as the cancel button, open the window without making any changes. 3. Data archiving window: No. 5 2 is not shown. Figure 5_2 shows the simulation results data archived by the lesser ', 歹^ input and output', in the "archive storage," can be carried out on the vessel, Aberdeen®, window The components in the description are the same as: 'Material input window. The #宏扪棕 case format restrictions and changes in the slot are: In the figure (7) slot type area, Λ ^ 埤 two different windows will be archived t format is restricted, the user's helmet must be...Bei Du ~ will store ascii's data file; Miscellaneous format. 19 201118764 4. Typhoon path selection window ^ When the data input is completed, you can select the upper tool column, The pattern calculation in the mode calculation is used to perform the mode estimation. This real mode automatically lists the 7 paths to provide the user with a choice. Please refer to Figure 5·3, Figure 5_3, '', typhoon road selection window'. The seven classifications are determined by the Hualien Port Authority in Table 2. The following is an example of the complete operation of the model: Available tools in the Matlab environment The slot open function of the column or the direct input of C_and WindGW is the name of the interface. The name of the block is currently tentatively set to mship.p (as shown in Figure 6). After entering the (4) dynamic warning mode GUI, I saw that there are four tiles in the main work area, as well as the warning light, but the simulation calculation cannot be displayed by the input data of ...s and Wo. In the upper toolbar, this sister, & Select, read the capital for the input of the bait, or select the "instant wave height and instant satellite imagery in the external Weidan m 来 to query the data" and the palace soil, and the Gan and the "he" The user manual,, and, copyright H has its options such as "data storage, and,, mode operation". This is to avoid the user's wrong fall = anti-white state can not be selected, report. In this step we can choose "Input and: fruit Read data. And according to Zhao Di, w into the shell material ', enter the window to select the input data. In addition, the default switchable name in the hard disk of the host computer or the network hard disk in the figure 6.3 is (*.dat), and the user can adjust the file name of the file. Type (") to find the name of any ascii file that you want to enter. 2011. The current typhoon data input format uses the simplest plain text file. The file contains the following typhoon data: I Temple Gate 歹丨 2. Year 3. Month 4.曰5.6.Longitude 7.Latitude 8·Wind speed 9·Zhao Feng alarm status, the format is not strictly limited, users can use EXCEL or document editing software to make this input file, in each column of data Tab, Space, comma, etc. can be used to indicate the interval. If there is any omission in the data, the operator needs to do the interpolation or addendum processing first. After the processing, adjust the time series (from the beginning to the last data) to start the input. • If the data length and format are not problematic after the typhoon data input is completed, the surface of the wind center, the maximum wind speed, the distance between the hurricane center and the Hualien port, and the typhoon path map will be automatically displayed in the window block of the interface. Figure 6_4 while viewing Photo after the issue can not be ,, click op mode in the toolbar projections after the "path selection to start the simulation projections. As shown in the figure, the situation after the completion of the calculation is shown. The lower right of the interface is the change of the ship's transaction index with time, and the interface is not marked by the warning light. After the calculation is completed, you can use the upper toolbar, and input and export the data in the "out" to save. In summary, the windowed operation mode of the ship dynamic early warning system not only provides the user with the convenience of operation, but also integrates the operation of the whole mode. The mode after the sigma can be used to improve the cooperation degree of the future automation environment. The future can be applied to provide online real-time ship dynamic mode estimation, which can directly extract typhoon observation data in an unmanned environment to calculate the ship's transaction index and simultaneously publish it with the server.
[S 21 201118764 【圖式簡單說明】 第1圖顯示颱風中心置 的示性波高關係圖。測站的距離與觀測站 貝示跑風風速與船隻異動指數關係圖。 I 員不颱風中心至測站方位 指數 關係圖。 又[S 21 201118764 [Simple description of the diagram] Figure 1 shows the relationship between the indicative wave heights of the typhoon center. The distance between the station and the observatory station shows the relationship between the wind speed and the ship's transaction index. I do not have a typhoon center to the station orientation index diagram. also
圖顯示影響船舶動態之颱風因子示意圖。 圖顯不類神經網路架構示意圖。 第4-1圖顯示介面主視窗。 第4-2圖顯示樹狀結構的工具列。 第4-3圖顯示船舶異動指數所對應之警示標德。 第5-1圖顯不颱風資料輸入視窗。 第5-2圖顯示模擬結果資料存檔視窗。 第5-3圖顯示颱風路徑選擇視窗。 第6-1圖顯示本發明—實施例之操作流程。 第6-2圖顯示本發明一實施例之操作流程。 第6-3圖顯示本發明一實施例之操作流程。 第6-4圖顯示本發明一實施例之操作流程。 第6-5圖顯示本發明—實施例之操作流程。 【主要元件符號說明】 (無元件符號說明) 22The figure shows a schematic diagram of the typhoon factor that affects ship dynamics. The diagram shows a schematic diagram of a neural network architecture. Figure 4-1 shows the main interface of the interface. Figure 4-2 shows the toolbar of the tree structure. Figure 4-3 shows the warning signs corresponding to the ship's transaction index. Figure 5-1 shows the typhoon data input window. Figure 5-2 shows the simulation results data archive window. Figure 5-3 shows the typhoon path selection window. Figure 6-1 shows the operational flow of the present invention - an embodiment. Fig. 6-2 shows the operational flow of an embodiment of the present invention. Fig. 6-3 shows the operational flow of an embodiment of the present invention. Fig. 6-4 shows the operational flow of an embodiment of the present invention. Figures 6-5 show the operational flow of the present invention - an embodiment. [Main component symbol description] (No component symbol description) 22
Claims (1)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW98140565A TW201118764A (en) | 2009-11-27 | 2009-11-27 | Ship dynamic mode of operation of early warning systems for windows |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW98140565A TW201118764A (en) | 2009-11-27 | 2009-11-27 | Ship dynamic mode of operation of early warning systems for windows |
Publications (1)
Publication Number | Publication Date |
---|---|
TW201118764A true TW201118764A (en) | 2011-06-01 |
Family
ID=44935843
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW98140565A TW201118764A (en) | 2009-11-27 | 2009-11-27 | Ship dynamic mode of operation of early warning systems for windows |
Country Status (1)
Country | Link |
---|---|
TW (1) | TW201118764A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2695808A1 (en) | 2012-08-10 | 2014-02-12 | ABB Research Ltd. | Required attentiveness indicator |
-
2009
- 2009-11-27 TW TW98140565A patent/TW201118764A/en unknown
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2695808A1 (en) | 2012-08-10 | 2014-02-12 | ABB Research Ltd. | Required attentiveness indicator |
WO2014023483A1 (en) | 2012-08-10 | 2014-02-13 | Abb Research Ltd | Attentiveness indicator |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bass et al. | Surrogate modeling of joint flood risk across coastal watersheds | |
Kim et al. | A time-dependent surrogate model for storm surge prediction based on an artificial neural network using high-fidelity synthetic hurricane modeling | |
CN111832810A (en) | Flood forecasting model, information processing method, storage medium, and computer device | |
Nederhoff et al. | Simulating synthetic tropical cyclone tracks for statistically reliable wind and pressure estimations | |
Won et al. | Development and application of an urban flood forecasting and warning process to reduce urban flood damage: A case study of Dorim River basin, Seoul | |
Hagen et al. | Development and evaluation of flood forecasting models for forecast-based financing using a novel model suitability matrix | |
CN115267945A (en) | Thunder and lightning early warning method and system based on graph neural network | |
Ávila et al. | Comparative evaluation of five hydrological models in a large-scale and tropical river basin | |
Wen et al. | Applying an artificial neural network to simulate and predict Chinese fir (Cunninghamia lanceolata) plantation carbon flux in subtropical China | |
Chondros et al. | A coastal flood early-warning system based on offshore sea state forecasts and artificial neural networks | |
Young | Flood risk assessment and forecasting for the Ganges-Brahmaputra-Meghna River basins | |
Lecacheux et al. | Toward the probabilistic forecasting of cyclone-induced marine flooding by overtopping at Reunion Island aided by a time-varying random-forest classification approach | |
Jafarzadegan et al. | Real-time coastal flood hazard assessment using DEM-based hydrogeomorphic classifiers | |
Banihabib | Performance of conceptual and black-box models in flood warning systems | |
Mar et al. | Optimum neural network architecture for precipitation prediction of Myanmar | |
Velasco et al. | Days-ahead water level forecasting using artificial neural networks for watersheds | |
Nkoana | Artificial neural network modelling of flood prediction and early warning | |
TW201118764A (en) | Ship dynamic mode of operation of early warning systems for windows | |
Lyddon et al. | Sensitivity of flood hazard and damage to modelling approaches | |
Zhao et al. | Intercomparison of empirical formulations of maximum wind radius in parametric tropical storm modeling over zhoushan archipelago | |
Meles et al. | Curating 62 years of walnut gulch experimental watershed data: Improving the quality of long-term rainfall and runoff datasets | |
Sharma et al. | Hydrologic simulation approach for El Niño Southern Oscillation (ENSO)-affected watershed with limited raingauge stations | |
Rashid et al. | Stochastic simulation of storm surge extremes along the contiguous United States coastlines using the max-stable process | |
Gu et al. | A Stacking Ensemble Learning Model for Monthly Rainfall Prediction in the Taihu Basin, China. Water 2022, 14, 492 | |
Nederhoff et al. | Accounting for Uncertainties in Forecasting Tropical Cyclone-Induced Compound Flooding |