1374393 九、發明說明： 【發明所屬之技術領域】 本發明是有關於一種使用飛航安全裕度理論之飛航操作 風險評估方法、風險評估專家系統及其建立方法，且特別是 有關於用以評估飛航操作之安全性的飛航安全裕度風險評估 方法、風險評估專家系統及其建立方法。 【先前技術】 目前國際民航界’常需大量地使用飛航操作品質保證系 統（FHght 〇perational Quality Assurance ; F〇QA)來紀錄飛機飛 行時之各項數據’再透過設定—些飛行限制範圍，以確保飛 航操作的品質。在風險評估上，則係依據主觀、且非量化的 風險矩陣，來協助飛安管理。 現有飛安驗評估的做法，是以風險矩料主^橫轴係 表不事件發生的或然率，其可區分為·非常頻繁、常常、偶 爾、很少、不太可能等五級。縱轴係表示事件的嚴重性，兑 可區分為·’毁滅性、嚴重、重大、輕微、可忽略等五級。再 針對每事件，綜合其或然率及其嚴重性，進行相對主觀的 然而 上遮飛女風險評估的方法具有一些缺點。第一個 缺點係過於主觀，-個事件的嚴重性，常常是見仁見智，因 人而異的’因而無法明確客觀地評估事件的嚴重性 缺點係不科學’由於主觀的認定，常常會因時、因地而：， 因而判斷標準隨時都在變動。第三個缺點係最關鍵的，以現 1374393 有作法而言，事件的錢率及其嚴重性只區分等級，而&法 進行數值量化，導致造成現㈣法完全無法針對飛航安全性 的變動，來提供有用的訊息。 此外，目前有關飛航安全問題的探討，大都係以管理觀 點出發，例如從序列式因果觀點來看，將飛安事故的發 程利用骨牌理論來說明，其中每_事件的發生有如骨牌— ，’引發下-階段事件的發生。因此，當飛行過程中的某環 節出現錯誤，即環環相扣成鍊而造成事故。然而，上述管理 理論無法進-步對事故風險的評估，進行量化因而無 立出具備有分析計算能力的工具。 再者，若從序列式因果觀點出發，自然地也會以序列式觀點 來看飛航風險，因而得出序列式的風險評估方式，即按照 航程序中，某·環節疏失，所可能造成危害的嚴重性，^ 個參數超限之嚴重性’來評估飛航之風險。然而，現代的民 航機，係-種大型的複雜系統’單純的疏失，可能_法 預料的複雜連鎖反應，在同，不同事件同時地在 例如’當飛機的外形結構遭到破壞’其可能同時造 統損害、漏油或電力系統損害。如此高度複雜以及緊人 的特性’使得序列式的飛安觀點，常常不足以完整： 航複雜狀況的風險〇 m 【發明内容】 因此，本發明之一方面，係在於提 诃货種使用飛航安全 袼度理狀飛織錢輯財法、驗料專㈣統= 2方法，藉以科學化地評估飛航之安全性，並提供量化評 估飛航操作安全性的科學化工具。 本發明之又H係在於提供-種使㈣航安全裕度 理論之飛航觸缝評財法、風險評料㈣統及其建立 方法’藉以根據事件的整體情境或過程，來評估飛航風險， 而非單單僅根據某些個別的參數。 本發明之又-方面，係在於提供—種使用飛航安全裕度 理論之飛㈣作騎評㈣法、恥評估專㈣統及其建立 方法，藉以呈現飛航過程中安全性的連續變化，因而可清楚 地分析飛安事件之任的安全性和異常因素。 根據本發明之實施例’本發明之風險評估方法係用以評 估飛航女全性，其中風險評估方法至少包含：建立複數個 訓練樣本，其巾建立每_此些樣本的步驟至少包含：由 專家根據㈣個預讀境參數，來評估在飛航絲中，避免 發生預《又事件情境，所需之综合能力並進行評分；計算 該些專家料分之分數，以轉—综合能力值；以及根據综 5月b力值| 4异得到_•預設之安全裕度，藉以根據此些預 設情境參數，與預設安全裕度之間的相對_，來建立每— 此些訓練樣本，矛lj用此些訓練樣本，來訓練一類神經網路； 以及在訓練類神經網路後’建立飛航情境與安全裕度的一般 因果關係，再輸人複數個情境參數於類神經網路，並利用類 神經網路，來推算出任-所予情境之安全裕度，藉以根據安 全裕度來汗估飛航安全性。 又，根據本發明之實施例，本發明之飛航安全裕度風險 13743931374393 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a flight operation risk assessment method, a risk assessment expert system and a method for establishing the same using the flight safety margin theory, and in particular to A flight safety margin risk assessment method, a risk assessment expert system, and a method for establishing the safety of the flight operation. [Prior Art] At present, the international civil aviation community often needs to use a large number of FHght 〇perational Quality Assurance (F〇QA) to record the data of the aircraft's flight, and then set the flight limits. To ensure the quality of the flight operation. In the risk assessment, it is based on a subjective and non-quantitative risk matrix to assist Fei'an management. The existing practice of flying safety assessment is based on the risk probability that the main axis and the horizontal axis show the probability of occurrence of events, which can be divided into five levels: very frequent, often, occasionally, rarely, and unlikely. The vertical axis indicates the severity of the incident, and the redemption can be divided into five levels: **destructive, serious, significant, minor, negligible. Then, for each event, the probability and its severity are combined to be relatively subjective. However, the method of risk assessment for females has some shortcomings. The first shortcoming is too subjective, the severity of an event is often a matter of opinion, and it is different from person to person. Therefore, it is impossible to clearly and objectively assess the seriousness of the event. The shortcoming is unscientific. Due to subjective identification, it often happens from time to time. Because of the ground: Therefore, the judgment standard is changing at any time. The third shortcoming is the most critical. In the case of the current 1743393 method, the money rate and its severity of the event are only classified, and the numerical method of the & method results in the fact that the current (four) method is completely incapable of ensuring the safety of the flight. Change to provide useful information. In addition, most of the current discussions on flight safety issues are based on management perspectives. For example, from the perspective of serial causality, the origin of the Fei'an accident is explained by the domino theory, in which each event occurs like a domino. 'Initiation of the next-stage event. Therefore, when a certain loop in the flight occurs, the loops are interlocked into a chain and cause an accident. However, the above management theory cannot further evaluate the risk of accidents and quantify them, thus eliminating the need for tools with analytical computing capabilities. Furthermore, from the point of view of sequence-based causality, the risk of flight will naturally be viewed from a serial perspective, and thus a sequential risk assessment method will be obtained, that is, in the course of the navigation procedure, the loss of a certain link may cause harm. The severity, the severity of the parameter overruns, is used to assess the risk of flight. However, modern civil aircraft, a large-scale complex system 'simple loss, may be expected to be a complex chain reaction, in the same time, different events at the same time, for example, 'when the aircraft's outer structure is destroyed' it may Damage caused by damage, oil spills or power system damage. Such a highly complex and tight-characteristic feature makes the sequence-based view of Fei'an often not complete: the risk of a complex situation 〇m [Summary] Therefore, one aspect of the present invention is to use the flight type The safety measure is based on the scientific evaluation of the safety of flight and provides a scientific tool to quantitatively evaluate the safety of flight operations. The present invention is based on the provision of the (four) navigation safety margin theory of the flight touch seam assessment method, the risk assessment (four) system and its establishment method 'by relying on the overall situation or process of the event to assess the flight risk , not just based on some individual parameters. A further aspect of the present invention is to provide a flight (four) for the use of the flight safety margin theory, a ride evaluation (four) method, a shame evaluation (four) system and a method for establishing the same, thereby presenting a continuous change in safety during flight. Therefore, the safety and abnormal factors of the Fei'an incident can be clearly analyzed. According to an embodiment of the present invention, the risk assessment method of the present invention is for evaluating flight female integrity, wherein the risk assessment method comprises at least: establishing a plurality of training samples, and the step of establishing each of the samples comprises at least: According to the (four) pre-reading parameters, the experts evaluate the pre-existing event situation, the comprehensive ability required and score in the flying air; calculate the scores of the experts, and transfer the comprehensive ability value; And according to the comprehensive force value of the May b | 4 different _ • preset safety margin, based on these relative situation parameters, and the relative safety margin between the preset _, to establish each of these training samples The spear lj uses these training samples to train a type of neural network; and after the training neural network, 'generates the causal relationship between the flight situation and the safety margin, and then enters multiple context parameters into the neural network. And use the neural network to estimate the safety margin of the situation, so as to estimate the safety of the flight according to the safety margin. Moreover, according to an embodiment of the present invention, the flight safety margin risk of the present invention is 1374393
=估專家系統，係用以評估飛航安全性，其中風險評估專家 …统’至少包含有使用者介面資料庫及推論單元。使用者 介面讀人複數個情境參數，並_驗評估專家系 統之運算結果。資料庫具有複數個訓練樣本，其中專家係根 據複數個預設情境參數，來評估在飛航過程中，避免發生一 預設事件情境’所需之綜合能力，並進行評分，而根據此些 專家所評分之純，來取得—综合能力值，㈣合能力值係 用以推算-預設之安全裕度（SafetyMa㈣每—訓練樣本， 係根據預設情境參數，與預設安純度之間的相對關係來建 立》推論單元設有-類神經網路，其中類神經網路係利用此 些訓練樣本’來進行訓練。其中，當輸人此些情境參數於類 神經網路時，類神經網路推算出—安全裕度，藉以根據安全 裕度來ff·估飛航安全性。 二又，根據本發明之實施例，本發明之飛航安全裕度風險 評估專家系統的建立方法，至少包含建立複數個訓練樣本， 其中建立每—此些訓練樣本的步驟，至少包含：由專家根據 複數個預設情境參數，來評估在飛航過程中，避免發生一預 設事件情境’所需之综合能力，並進行評分；計算該些專家 所評分之分數，以取得一综合能力值；以及根據综合能力值， 來冲算得到一預設安全裕度（Safety Margin)，藉以根據此些預 5又情境參數，與預設安全裕度之間的相對關係，來建立每一 此些訓練樣本；提供一資料庫，並儲存該些訓練樣本於該資 料庫中，提供一推論單元，其中推論單元設有一類神經網路； 利用此些訓練樣本，來訓練類神經網路；以及提供一使用者 1374393 介面，用以輸人複數個情境參數’於類神經網 類神經網路所推算之一安全裕度。 ’’ —因此，本發明之飛航安全㈣風时估方法、風險評估 =家系統及其建立方法，可藉由科學化、客觀及數據化的方 式’來呈現飛航安全狀況的連續變化，因而可清楚地分析飛 二事件之心㈣时全料異常㈣且可提㈣安風險 砰估的可靠性、正確性及全面性。 【實施方式】 為讓本發明之上述和其他目的、特徵、優點與實施例能 ' 更明顯易懂’本說明書將特舉出-系列實施例來加以說明。 • 但值得注意較，此些實施例只是用以說明本發明之實施方 式，而非用以限定本發明。 —請參照第1 ®，其繪示依照本發明之實施例之風險評估 專家系.统的系、统方塊圖。本實施例之飛航安全裕度風險評估 Φ 方法與風險評估專家系統100係用以評估飛航安全性，並可 做為協助業界在飛航安全管理、風險評估或趨勢預測之評估 方法和工具。本實施例之風險評估專家系統100可包含使用 者介面110、資料庫120、推論單元130、發展者介面14〇及 系統介面150。使用者介面110例如為鍵盤和顯示裝置之組 合，用以輸入複數個情境參數於專家系統1〇〇中，並可顯示 此專家系統100之運算結果。資料庫120例如為電腦裝置或 記憶裝置（例如硬磉或記憶體），用以儲存複數個訓練樣本 (Training pattern) ’其中訓練樣本係由專家來協助建立。推論 1374393 單元m較佳為電腦裝置，其連接於使用者介φ ιι〇和資 120’其中推論單凡13〇設有_類神經網路i3i，（例如類神經 網路軟體）’類神經網路131可藉由訓練樣本來進行訓練，藉 以數據化地推論飛航操作的安全性。發展者介& Μ係連^ 於資料庫120和推論單元13G，用以輸人_樣本於資料庫 ⑽令，並可對倾庫12()或推論單％丨料行編缉與修訂。 系統介面！50係連接於推論單以3〇肖一外部系統或設備（未 繪不）之間，藉以整合本實施例之風險評估專家系統1〇〇與其 他系統或設備’例如系統介面15G可連接於飛行裝置（例如飛 機或直升機）’而可作為一即時警示系統，以即時警示當時之 飛航安全性，因而可具有飛安風險的即時監控能力，提供駕 駛員多一套操作安全參考的工具。同時，本實施例之風險評 估專家系統100，亦可單獨用以做為管理決策輔助工具。 值得注意的是’在一實施例中’使用者介面11 〇、發展者 介面140及系統介面150可以係共用於一相同的介面裝置。 請參照第2圖，其繪示依照本發明之實施例之風險評估 方法的方法流程圖◊本實施例之飛航安全裕度風險評估方法 可包含有建立複數個訓練樣本（步驟1〇1);利用訓練樣本，來 訓練類神經網路131(步驟1〇2);以及在訓練類神經網路131 後’輸入複數個情境參數於類神經網路131，並利用此類神經 碼路13丨來推算出一安全裕度（Safety Margin)(步驟103)，藉 以根據此安全裕度，來評估飛航的操作安全性。值得注意的 义’以下實施例之敘述係以飛機的飛航安全評估來舉例說 明’然不限於此’本實施例之風險評估方法與風險評估專家 393 系統100亦可應用於其他飛行交通工具，例如直升機。 人請參照第3 ® ’其繪示依照本發明之實施例之安全裕度 概^圖。本發明風險評估方法與風險評估專㈣統係、利用飛 航安王裕度（Flight Safety Margin)理論來進行評估，以下進一 步對飛航安全裕度加以說明。首先，可^義—情衫間，其 ^表所有會對飛航安全造成影響的因素所成的集合。此情境 空間中的任—點係代表影響飛航安全之因素的某種組合。因 此，任何-趟飛行，均可时境Μ中之—條連續曲線來代 表。、通常，在任何-個飛行任務中，皆具有—標準的理想情 境（以中心線201來表示），此理想情境會隨不同之飛行階段而 有不同，且會隨時間連續變動。此外，在情境空間中，當所 =影響飛安的因素形成某種特定組合時，某種飛航事件即可 能發生’而此事件情境（或事件情形）也會隨不同飛行階段而有 所不同’在本實施财，此事件情境係以事件邊界2() 表。 在理想條件下，飛行過程應沿著中心線2〇1進行然而， 實際條件與理想條件之間，必然存在有落差1此，實際飛 行的情境，必料沿著中心線2G1附近進行。在本實施例令， 任何欲探討飛行操作安全性的—瞬間（時間點），可藉由當下情 境2〇3來表示。因此，在此情境空間申’任一當下情境203 與事件邊界2G2之間的距離，即可表*為#下情境距離事件 的安全裕度°其中飛航安全裕度，可代表飛航組員操作飛機 的安全空間。在理想條件下，飛行過程的當下情境2们係沿 中心線201進行，直到完成飛航任務。然而，實際上若發生 1374393 丨人為疏< 機械故障或天候因素等影響，飛行過程的 下情境203將離中心線201更遠。因此，安全裕度的變動可 呈現人為疏失或其它各種因素所造成的風險因而可用以評 2行過程中的操作安全性。其中事件邊界2Q2可以係預設 為t示任意的事件情境，例如：飛機墜毁、異常事件、穩定 進場、安全落地、衝出跑道等。The evaluation expert system is used to assess the safety of the flight. The risk assessment expert has at least a user interface database and an inference unit. The user interface reads a plurality of context parameters and evaluates the results of the evaluation of the expert system. The database has a plurality of training samples, wherein the expert evaluates the comprehensive ability required to avoid a predetermined event situation during the flight process according to a plurality of preset context parameters, and performs scoring according to the experts. The score is pure, to obtain the comprehensive ability value, and (4) the ability value is used to calculate the preset safety margin (SafetyMa (four) per-training sample, based on the relative situational parameters, and the relative safety between the preset The relationship is established. The inference unit is provided with a neural network, in which the neural network is trained by using these training samples. Among them, when the input of these context parameters is in the neural network, the neural network Deriving a safety margin to estimate the flight safety based on the safety margin. Secondly, according to an embodiment of the present invention, the method for establishing a flight safety margin risk assessment expert system of the present invention includes at least establishing a plurality of training samples, wherein the step of establishing each of the training samples comprises at least: evaluating, by the expert, a plurality of preset context parameters during flight, Exempt from the pre-set event situation's required comprehensive ability and score; calculate the scores scored by the experts to obtain a comprehensive ability value; and calculate a preset safety margin based on the comprehensive ability value (Safety Margin), based on the relative relationship between the pre-five context parameters and the preset safety margin, to establish each of the training samples; providing a database and storing the training samples in the data In the library, an inference unit is provided, wherein the inference unit is provided with a type of neural network; the training samples are used to train the neural network; and a user 1734393 interface is provided for inputting a plurality of context parameters 'in the class A safety margin calculated by a neural network-like neural network. '' - Therefore, the flight safety (four) wind time estimation method, risk assessment = home system and its establishment method of the present invention can be scientific, objective and data The way of 'to provide a continuous change in the safety status of the flight, so that the heart of the second event can be clearly analyzed (4) when the total anomaly (4) and can be raised (4) the reliability of the risk assessment The above and other objects, features, advantages and embodiments of the present invention will become more <RTIgt; <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; It is to be noted that the embodiments are merely illustrative of the embodiments of the invention and are not intended to limit the invention. - Referring to FIG. 1 , which illustrates a risk assessment expert system in accordance with an embodiment of the present invention. System and system block diagram. The flight safety margin risk assessment Φ method and risk assessment expert system 100 of this embodiment is used to assess flight safety and can be used to assist the industry in flight safety management, risk assessment or trend. The estimated risk assessment expert system 100 can include a user interface 110, a database 120, an inference unit 130, a developer interface 14 and a system interface 150. The user interface 110 is, for example, a combination of a keyboard and a display device for inputting a plurality of context parameters in the expert system 1 and displaying the result of the operation of the expert system 100. The database 120 is, for example, a computer device or a memory device (e.g., a hard memory or a memory) for storing a plurality of training patterns, wherein the training samples are assisted by an expert. It is inferred that the 1744393 unit m is preferably a computer device connected to the user medium φ ιι〇 and the capital 120', wherein the inference of the singular 13 〇 is provided with a _ class neural network i3i, (for example, a neural network software) The road 131 can be trained by training samples to logically infer the safety of the flight operation. Developers & Μ ^ 于 in database 120 and inference unit 13G, used to input _ samples in the database (10) order, and can edit and revise the dumping 12 () or inferences. System interface! The 50 series is connected between the inference list and the external system or equipment (not shown), so as to integrate the risk assessment expert system of the present embodiment, and other systems or devices, such as the system interface 15G, can be connected to the flight. The device (such as an airplane or helicopter) can be used as a real-time warning system to instantly alert the flight safety at that time, so that it can have the real-time monitoring capability of the risk of flying security, and provide the driver with a set of tools for operating safety reference. At the same time, the risk assessment expert system 100 of the present embodiment can also be used alone as a management decision aid. It is noted that in one embodiment, the user interface 11, the developer interface 140, and the system interface 150 may be used in common for one and the same interface device. Please refer to FIG. 2, which illustrates a method flowchart of a risk assessment method according to an embodiment of the present invention. The flight safety margin risk assessment method of the present embodiment may include establishing a plurality of training samples (step 1〇1). Training the neural network 131 using the training samples (steps 1 and 2); and inputting a plurality of context parameters to the neural network 131 after training the neural network 131, and using such neural code 13 To calculate a safety margin (step 103), based on this safety margin, to assess the operational safety of the flight. It is noted that the following description of the embodiments is exemplified by the flight safety assessment of the aircraft. However, the risk assessment method and risk assessment expert 393 system 100 of the present embodiment can also be applied to other flight vehicles. Such as a helicopter. Please refer to Section 3®' for a safety margin profile in accordance with an embodiment of the present invention. The risk assessment method and the risk assessment (4) of the present invention are evaluated by the theory of Flight Safety Margin, and the flight safety margin is further explained below. First of all, it can be used to determine the set of factors that affect the safety of the flight. The arbitrarily-point in this situational space represents some combination of factors that affect flight safety. Therefore, any -趟 flight can be represented by a continuous curve in time. Usually, in any mission, there is a standard ideal situation (represented by centerline 201), which will vary from flight to flight and will change continuously over time. In addition, in the situational space, when the factors affecting Fei'an form a certain combination, some kind of flight event may occur' and the event situation (or event situation) will also vary with different flight phases. 'In this implementation, this event situation is based on the event boundary 2 () table. Under ideal conditions, the flight process should be carried out along the centerline 2〇1. However, there must be a gap between the actual conditions and the ideal conditions. The actual flight situation must be carried out along the centerline 2G1. In the present embodiment, any instant (time point) for discussing the safety of flight operations can be represented by the current situation 2〇3. Therefore, in this situation space, the distance between any current situation 203 and the event boundary 2G2 can be expressed as the safety margin of the situational distance event. The flight safety margin can represent the operation of the flight crew. The safe space of the aircraft. Under ideal conditions, the current situation 2 of the flight process is carried out along the centerline 201 until the flight mission is completed. However, in the event of a 1374393 丨 artificial spalling, mechanical failure or weather factor, the lower scenario 203 of the flight process will be further away from the centerline 201. Therefore, changes in safety margin can present risks due to human error or other various factors and can therefore be used to assess operational safety during the course of the line. The event boundary 2Q2 can be preset to indicate any event situation, such as: aircraft crash, abnormal event, stable approach, safe landing, rushing out of the runway, etc.
本實施例之訓練樣本係用以訓練類神經網路⑶，藉以 類神經_ 131學# —當下情境加⑽如麟情境)與其安 全拾度的因果關係。訓練樣本係由多位專家（例如專業駕幻 來，助建立’藉以吸取專家的知識與經驗，並透過專家對飛 航安全的認知’來建立驗評估專家线⑽。线 树’首先，由專家根據複數個預設情境參數，來評估當操 作交通卫具時’避免發生1設事件情境（㈣事件邊界加） 所需之綜合能力’並進行評分。接著，計算此些專家所評分 之分數’以取得-综合能力值1分的方式及標準，隨著所 欲探討安全性之事件而異，且係透過與專家之深度訪談來確 定，經取得所需之綜合能力值之後，經過倒數，即可得安全 裕度值。 以飛航安全為例，由當下情境203飛回標準正常情境， 所需之綜合飛行能力的大小’可用以代表飛行事件的嚴重 性’亦即為飛航安全裕度之大小。若當下情境2〇3至中心線 201的距離越遠，離事件邊界2〇2的距離即越近，則表示情況 越嚴重，飛航組員也需越高超的能力由當下情境2〇3飛二正 常情境。因此’可對多位資深飛行員（專家）進行訪談提供專 12 1374393 家在當下情境203時的預設情境參數，透過生理（例如臨場反 應與飛行技能）和心理（例如經驗與知識）等因素，以定義綜合 飛行能力，並由專家進行評分，藉以由專家來提供一客觀且 量化的分數。接著，收集和計算（例如平均）此些專家所評分之 分數，因而可取得一綜合能力值。其中，此些預設情境參數， 包含軟體參數（例如：飛行程序、標準或規定等）、硬體參數（例 如：飛機的高度、速度或攻角）及環境參數（例如：氣象、機場 條件或機場管制）等情境參數。 請參照第4圖，其繪示依照本發明之實施例之综合能力 問卷的示意圖。舉例來說，在本實施例中，每一訓練樣本， 可採用特定航班為範例，在其飛行過程中，選取複數個瞬間（時 間點），並設定一預設事故情形，例如本實施例為飛機墜毀事 故。接著，可參考 FOQA (Flight Operation Quality Assurance) 系統與國際飛安基金會所建議之ALAR (Approach- and-Landing Accident Reduction)中，提供的飛行員須注意的項 目，來製造綜合飛行能力問卷，其中此綜合飛行能力問卷係 預先區分綜合飛行能力的等級（例如高、中、低）。接著，訪談 資深飛行員（專家），並提供飛行員在特定航班中之某一瞬間的 複數個情境參數（預設情境參數）。接著，由飛行員根據在此瞬 間的情境參數，來評估出在飛行時避免發生預設事故情形（飛 機墜毀）所需之綜合能力，並對综合飛行能力問卷進行回答’ 以進行評分，飛行員之評估，可依據综合能力的所需程度或 等級來進行評分，藉以定義和量化綜合能力。接著，在收集 和計算多位專家的綜合飛行能力問卷後，依專家意見進行計 13 1374393 算，以取得綜合能力值（p)。 然後，根據綜合能力值（p)，來計算得到一預設安全裕度。 在本實施例中，預設安全裕度（SM)可由下列公式來計算得到：The training samples of this embodiment are used to train a neural network (3), and the causal relationship between the neural network and the safety of the current situation is increased by (10). The training sample is established by a number of experts (such as professional driving, to help build the 'knowledge and experience of the experts, and through the expert's understanding of flight safety' to establish an evaluation expert line (10). Line tree' first, by experts Based on a plurality of preset context parameters, evaluate the 'comprehensive ability to avoid the occurrence of a set event context ((4) event boundary plus) when performing a traffic aid and score. Then, calculate the scores of these experts' scores' The method and criteria for obtaining a comprehensive ability value of 1 point vary with the event of the security to be explored, and are determined through in-depth interviews with experts. After obtaining the required comprehensive ability value, after the countdown, The safety margin value can be obtained. Taking flight safety as an example, the current situation 203 is returned to the standard normal situation, and the required comprehensive flight capability 'can be used to represent the severity of the flight event' is also the flight safety margin. If the distance from the current situation 2〇3 to the centerline 201 is farther, the closer the distance from the event boundary 2〇2 is, the more serious the situation is, the higher the flight crew member needs to be. The ability to fly from the current situation 2〇3 to the second normal situation. Therefore, 'interview with a number of senior pilots (experts) can provide 12,314,393 preset situational parameters in the current situation 203, through physiology (such as on-site reaction and flight Factors such as skills) and psychology (such as experience and knowledge) to define integrated flight capabilities and scored by experts to provide an objective and quantified score by experts. Next, collect and calculate (eg, average) such experts The score of the score, thus obtaining a comprehensive ability value, wherein the preset context parameters include software parameters (eg, flight procedures, standards or regulations, etc.), hardware parameters (eg, aircraft altitude, speed or angle of attack) And context parameters such as environmental parameters (eg, weather, airport conditions, or airport control). Referring to Figure 4, a schematic diagram of a comprehensive competency questionnaire in accordance with an embodiment of the present invention is shown. For example, in this embodiment For each training sample, a specific flight can be used as an example. During the flight, multiple moments are selected (time) And set a preset accident situation, for example, this is a plane crash accident. Then, refer to the FOQA (Flight Operation Quality Assurance) system and the ALAR (Approach-and-Landing Accident Reduction) recommended by the International Fei'an Foundation. Provide the pilot's project to pay attention to the comprehensive flight capability questionnaire, which is a pre-division of the level of integrated flight capability (eg, high, medium, and low). Then, interview senior pilots (experts) and provide A plurality of context parameters (pre-set situation parameters) of the pilot at a certain moment in a particular flight. Then, based on the situational parameters at this instant, the pilot evaluates the occurrence of a preset accident (flight crash) during flight. The comprehensive ability required and the answer to the comprehensive flight capability questionnaire 'for scoring, the pilot's assessment can be scored according to the required level or level of comprehensive ability, in order to define and quantify the comprehensive ability. Then, after collecting and calculating the comprehensive flight capability questionnaires of multiple experts, the calculation results are based on the expert opinion to obtain the comprehensive ability value (p). Then, a preset safety margin is calculated based on the comprehensive ability value (p). In this embodiment, the preset safety margin (SM) can be calculated by the following formula:
SM=1/P 值得注意的是，本實施例之預設安全裕度的計算方式僅 為一示範例，然不限於此，由於安全裕度係用以表示操作交 通工具，相對於某一事件所擁有的安全空間，其可為一相對 值或一標準化數值，用以對不同的事件（例如正常落地與飛安 事故）進行比較和評估。因此，亦可利用其他計算方式，來計 算综合能力值（P)，以求得預設安全裕度（SM)。 因此，在某一瞬間（時間點）的預設情境參數與其對應的預 設安全裕度之間的相對關係，即建立為一訓練樣本，亦即每 一訓練樣本具有一組輸入輸出值，訓練樣本的輸入為預設情 境參數，而訓練樣本的輸出為預設安全裕度。接著，重複上 述步驟，來建立多個訓練樣本（例如數百個），藉以提升類神經 網路13 1的學習能力。 在建立訓練樣本後，接著，利用訓練樣本，來訓練類神 經網路131，藉以使類神經網路131學習預設情境參數與其對 應的預設安全裕度之間的相對關係，因而學習完成後之類神 經網路131，可對任意的情境參數進行分析，並推論得到其對 應的安全裕度。其中此類神經網路13 1，可例如為多層次網路 (Multilayer Network)、霍普菲爾網路（Hopfield Network)、或 徑基函數網路（Radial Basis Function Network)、或支援向量網 路（Support Vector Machines )、或委員會網路（Committee 14 1374393SM=1/P It is worth noting that the calculation method of the preset safety margin of this embodiment is only an exemplary example, but is not limited thereto, since the safety margin is used to indicate that the vehicle is operated, relative to an event. The safe space that is owned can be a relative value or a standardized value for comparing and evaluating different events (such as normal landing and Fei'an accident). Therefore, other calculation methods can be used to calculate the comprehensive capability value (P) to obtain the preset safety margin (SM). Therefore, the relative relationship between the preset context parameter at a certain instant (time point) and its corresponding preset safety margin is established as a training sample, that is, each training sample has a set of input and output values, and training The input of the sample is a preset context parameter, and the output of the training sample is a preset safety margin. Next, the above steps are repeated to establish a plurality of training samples (e.g., hundreds) to enhance the learning ability of the neural network 13.1. After the training sample is established, the training neural network 131 is trained by using the training sample, so that the neural network 131 learns the relative relationship between the preset context parameter and its corresponding preset safety margin, and thus after learning is completed. The neural network 131 can analyze any situational parameters and infer the corresponding safety margin. Such a neural network 13 1 may be, for example, a Multilayer Network, a Hopfield Network, or a Radial Basis Function Network, or a support vector network. (Support Vector Machines), or committee network (Committee 14 1374393
Machines)。 請參照第5圖，其繪示依照本發明之實施例之類神經網 路與安全裕度的關係圖。在訓練推論單元130之類神經網路 131(類神經網路131學習完成）後，類神經網路13丨可用以對 任意的情境參數進行分析。當評估飛航安全性時，可藉由使 用者介面110,輸入某一瞬間的相關情境參數於類神經網路 131，類神經網路131即可根據此些情境參數’推算出其對應 之安全裕度’藉以評估飛航安全性和風險。在飛航過程中， 不同的瞬間（或時間點）上可能因人為疏失而有不同的情境變 化，亦即產生複數個情境參數（情境參數！、2…N)，而此些情 境參數可輸入訓練推論單元13〇之類神經網路131，類神經網 路131即可輸出一安全裕度值，以作飛航安全性的評估依據。 本實施例之風險評估方法與風險評估專家系統1〇〇，亦可 用以呈現在飛航過程（例如航班）中之相關安全性的連續變化 和變動過程，藉以評估其事件（操作過程）的異常狀況或事故發 生原因。當評估飛航安全性時，可擷取在飛航過程中的複數 個時間點（瞬間），接著，輸入在每一時間點上之相關的情境參 數於類神經網路131 ’藉以利用類神經網路丨31來推算在此飛 航過程中每一時間點上的安全裕度。因此，此些時間點上的 安全裕度可形成一安全裕度曲線，其可對應於飛航過程的時 間，用以呈現此飛航過程中之安全性的連續變化。 請參照第6圖，其繪示依照本發明之實施例之正常航班 與大霧航班的飛航安全裕度變化圖。以飛航安全裕度為例， 在本實施例中，飛航安全裕度係定義介於〇〜丨之間。在最標 15 1374393 準的理想情境下操作時，其安全性等於卜而代表百分之百安 全。當安全裕度被壓縮為Q時，代表飛行員所需的综合飛行 能力為無窮大’亦即發生事故，其安全性等於〇。以正常航班 與大霧航班來進行比較，對正常航班而言，落地時之平均飛 航安全裕度為0.531。而對大霧航班而言，由於最後落地時的 能見度受到影響，雖仍可安全落地，但其安全裕度由〇 531降 至0.483，因而充份顯示大霧對飛航安全裕度的影響程度。 請參照第7圖和第8圖，帛7圖係繪示依照本發明之實 施例之正常航班與名古屋空難的飛航安全裕度變化圖，第8 圖係緣示依照本發明之實施例之正常航班，與大園空難的飛 航安全裕度變化圖。以名古屋空難和大園空難事件為例將 名古屋事件最後2400呎，到失事點的所有情境參數輸入於類 神經網路131 ’可得到名古屋空難在最後258秒内的安全裕度 變化曲線，亦即飛機在最後258秒内的安全性變化。同樣地， 將大園空難最後3000呎，到失事點的所有情境參數，輸入於 類神經網路131 ’可得到大園空難在最後151秒内的安全裕度 曲線，亦即飛機在最後151秒内的安全性變化.因此，本實 施例之飛航安全裕度風險評估方法與風險評估專家系統 100，可清楚地以數據化的方式，呈現在飛安事件（操作過程） 中之安全性的連續變化，甚至到墜毀的情境。 值得注意的是，在本實施例中，飛航安全裕度風險評估 方法與風險評估專家系統1〇〇,可呈現安全裕度在時間軸上的 變化情形（安全裕度曲線）’然不限於此，亦可用以呈現安全裕 度在其他情境參數上的變化情形。例如，安全裕度可對應飛 1374393 :的高度或速度的變化，攻角或引擎轉速的變化藉以評估 乂通工具之操作對飛航安全性的影響。 月’’”、第9圖’其緣不依照本發明之實施例之風險評估 ^系統的建立方法流程圖。當建立本實施例之風險評估專 豕系統時’首先，建立複數個訓練樣本（步驟3〇1)。接著提 供資料庫12G，並儲存訓練樣本於資料庫12()中（㈣狗。 =著:提供推論單元13G(㈣3G3)，並利用訓練樣本來訓練 隹論早το 130之類神經網路131(步驟3〇4)。接著提供使用 者；I面11G(步驟3G5) ’用讀人情境參數於類神經網路⑶ 中，並可顯示類神經網路131所推算之安全裕度。 由上述本發明的實施例可知，本發明之飛航安全裕度風 險評枯方法、風輯估專㈣統及其建立方法可提供一個 數據化的飛安風險評估方法和系統，以協助飛航駕駛者或相 ^業者來提升飛航安全㈣的技術水準。且相較於傳統的飛 安管理理論，本發明之㈣安全裕度風崎估方法與風險評 估專家mx完全科學化、客觀及數據化的方式，來呈 現飛航安全狀況的連續變化，因而可清楚地分析飛安事件之 任一瞬間的安全性和影響因素。再者，由於本㈣之飛航安 全裕度風險評估方法，與風險評估專家线，係考慮飛安事 件的整體情境來進行分㈣評估’因而可提升㈣風險評估 的正確性與全面性。 雖然本發明已以實施例揭露如上，然其並非用以限定本 發明’任何熟習此技藝者，在不脫離本發明之精神和範圍内， 當可作各種之更動與潤飾，因此本發明之保護_當視㈣ 1374393 之申請專利範圍所界定者為準。 【圖式簡單說明】 為讓本發明之上述和其他目的、特徵、優點與實施例能 更明顯易懂，所附圖式之詳細說明如下： 第1圖係繪示依照本發明之實施例之風險評估專家系統 的系統方塊圖^ 、 第2圖係繪示依照本發明之實施例之風險評估方法的方 法流程圖。 第3圖係繪示依照本發明之實施例之安全裕度概念圖。 第4圖係繪示依照本發明之實施例之綜合能力問卷的示 意圖。 第5圖係繪示依照本發明之實施例之類神經網路與安全 裕度的關係圖。 ' 第6圖係繪示依照本發明之實施例之正常航班與大霧航 班的飛航安全裕度變化圖。 第7圖係繪示依照本發明之實施例之正常航班與名古屋 空難的飛航安全裕度變化圖。 第8圖係繪示依照本發明之實施例之正常航班與大園空 難的飛航安全裕度變化圖。 第9圖係繪示依照本發明之實施例之風險評估專家系統 的建立方法流程圖。 【主要元件符號說明】 1374393 100:風險評估專家系統 120 :資料庫 13 0 :推論單元 140 :發展者介面 201 :中心線 203 :當下情境 110.使用者介面 131 :類神經網路 15 0 ·系統介面 202 .事件邊界Machines). Referring to Figure 5, there is shown a relationship between a neural network and a safety margin in accordance with an embodiment of the present invention. After the neural network 131 (the neural network 131 learning is completed), such as the training inference unit 130, the neural network 13 can be used to analyze any context parameters. When assessing the safety of the flight, the user interface 110 can be used to input the relevant context parameters of a certain moment to the neural network 131, and the neural network 131 can calculate the corresponding security according to the context parameters. Yudu' is used to assess flight safety and risk. During the flight process, different moments (or time points) may have different situational changes due to human error, that is, multiple context parameters (situation parameters!, 2...N) are generated, and these context parameters can be input. The neural network 131, such as the training inference unit 13, can output a safety margin value for the assessment of flight safety. The risk assessment method and the risk assessment expert system of the present embodiment can also be used to present the continuous change and change process of the relevant security in the flight process (such as a flight), thereby evaluating the abnormality of the event (operation process). The cause of the situation or the accident. When assessing flight safety, multiple time points (instantaneous) during the flight can be retrieved, and then the relevant context parameters at each time point are input to the neural network 131' The network 丨 31 is used to estimate the safety margin at each point in the flight. Therefore, the safety margin at these points in time can form a safety margin curve that can correspond to the time of the flight process to present a continuous change in safety during the flight. Please refer to FIG. 6, which illustrates a flight safety margin change diagram of a normal flight and a foggy flight in accordance with an embodiment of the present invention. Taking the flight safety margin as an example, in this embodiment, the flight safety margin is defined between 〇~丨. When operating in the ideal situation of the standard 15 1374393, its safety is equal to b and represents 100% safety. When the safety margin is compressed to Q, the integrated flight capability required to represent the pilot is infinity', that is, an accident occurs, and its safety is equal to 〇. Comparing normal flights with heavy fog flights, the average flight safety margin for landing is 0.531 for normal flights. For heavy fog flights, the visibility is affected by the final landing, although it can still be safely landed, but its safety margin is reduced from 〇531 to 0.483, thus fully showing the impact of heavy fog on flight safety margin. . Please refer to FIG. 7 and FIG. 8 , FIG. 7 is a diagram showing a flight safety margin change diagram of a normal flight and a Nagoya air crash according to an embodiment of the present invention, and FIG. 8 is a diagram showing an embodiment according to the present invention. Normal flight, flight safety margin change chart with the big park air crash. Taking the Nagoya air crash and the big garden crash as an example, the last 2400 名 of the Nagoya incident, all the situation parameters to the crash point were input to the neural network 131 'The safety margin curve of the Nagoya air crash in the last 258 seconds, that is, the aircraft Security changes in the last 258 seconds. Similarly, the final 3000 大 of the big garden crash, all the situation parameters to the crash point, input to the neural network 131 ' can get the safety margin curve of the big garden air crash in the last 151 seconds, that is, the aircraft in the last 151 seconds Security change. Therefore, the flight safety margin risk assessment method and risk assessment expert system 100 of the present embodiment can clearly present the continuous change of safety in the Fei'an event (operation process) in a data-based manner. Even to the crashed situation. It is worth noting that in the present embodiment, the flight safety margin risk assessment method and the risk assessment expert system can present a change in the safety margin on the time axis (safety margin curve). This can also be used to present changes in safety margins over other context parameters. For example, the safety margin can correspond to the change in altitude or speed of the fly 1374393: the angle of attack or the change in engine speed to evaluate the impact of the operation of the tool on flight safety. Month'', Fig. 9 is a flowchart of a method for establishing a risk assessment system according to an embodiment of the present invention. When establishing the risk assessment system of the present embodiment, 'first, a plurality of training samples are established ( Step 3〇1). Then provide the database 12G, and store the training samples in the database 12() ((4) dogs. =: provide the inference unit 13G ((4) 3G3), and use the training samples to train the public opinion early το 130 Neural network 131 (step 3〇4). Then provide the user; I face 11G (step 3G5) 'Use the read context parameter in the neural network (3), and display the security of the neural network 131 According to the embodiment of the present invention, the method for assessing the safety margin of the flight safety of the present invention, the method for calculating the wind safety estimate, and the method for establishing the same can provide a data-based Fei'an risk assessment method and system, Assist the pilot of the flight or the industry to improve the technical level of flight safety (4). Compared with the traditional Fei'an management theory, the safety margin of the invention (4) safety assessment method and risk assessment expert mx is completely scientific and objective. And the data side To present a continuous change in the safety status of the flight, so that the safety and influencing factors of the Fei'an incident can be clearly analyzed. Furthermore, due to the risk assessment method of the flight safety margin of this (4), and the risk assessment expert Line, taking into account the overall situation of the Fei'an incident to carry out the sub-(four) evaluation 'Therefore, the correctness and comprehensiveness of the (4) risk assessment can be improved. Although the present invention has been disclosed above by way of example, it is not intended to limit the invention to any familiarity. The skilled person can make various changes and refinements without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention is defined by the scope of the patent application of the Japanese Patent Application No. 1374393. The above and other objects, features, advantages and embodiments of the present invention will become more <RTIgt; FIG. 2 is a flow chart showing a method of risk assessment according to an embodiment of the present invention. FIG. 3 is a diagram showing an embodiment of the present invention. Safety margin concept map. Fig. 4 is a schematic diagram showing a comprehensive capability questionnaire according to an embodiment of the present invention. Fig. 5 is a diagram showing a relationship between a neural network and a safety margin according to an embodiment of the present invention. Figure 6 is a diagram showing the flight safety margin change of a normal flight and a foggy flight in accordance with an embodiment of the present invention. Figure 7 is a diagram showing normal flight and Nagoya air crash flight in accordance with an embodiment of the present invention. FIG. 8 is a diagram showing a flight safety margin change of a normal flight and a large garden air crash according to an embodiment of the present invention. FIG. 9 is a diagram showing a risk assessment expert system according to an embodiment of the present invention. Flow chart of the establishment method. [Main component symbol description] 1374393 100: Risk assessment expert system 120: database 13 0: inference unit 140: developer interface 201: center line 203: current situation 110. user interface 131: neuron Network 15 0 · System Interface 202. Event Boundary
101 .建立複數個訓練樣本 102 .利用到練樣本來訓練類神經網路 1〇3 :輸人複數個情境參數於類神經網路 網珞來推算出安全裕度 301 :建立複數個訓練樣本 302 .提供資料庫，並儲存訓練樣本於資 303 :提供推論單元 ’並利用類神經 料庫中 3〇4 .利用訓練樣本來訓練類神經網路 305 :提供使用者介面101. Establishing a plurality of training samples 102. Using the training samples to train the neural network 1〇3: inputting a plurality of context parameters to the neural network to estimate the safety margin 301: establishing a plurality of training samples 302 Provide a database and store the training sample at 303: Provide the inference unit 'and use the neuron library 3. 4 . Use the training sample to train the neural network 305: Provide the user interface