JP2009059301A - Method for determining automatic termination of monte carlo evaluation - Google Patents

Method for determining automatic termination of monte carlo evaluation Download PDF

Info

Publication number
JP2009059301A
JP2009059301A JP2007228079A JP2007228079A JP2009059301A JP 2009059301 A JP2009059301 A JP 2009059301A JP 2007228079 A JP2007228079 A JP 2007228079A JP 2007228079 A JP2007228079 A JP 2007228079A JP 2009059301 A JP2009059301 A JP 2009059301A
Authority
JP
Japan
Prior art keywords
evaluation
monte carlo
value
confidence interval
limit value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
JP2007228079A
Other languages
Japanese (ja)
Inventor
Toshikazu Motoda
敏和 元田
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Japan Aerospace Exploration Agency JAXA
Original Assignee
Japan Aerospace Exploration Agency JAXA
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Japan Aerospace Exploration Agency JAXA filed Critical Japan Aerospace Exploration Agency JAXA
Priority to JP2007228079A priority Critical patent/JP2009059301A/en
Publication of JP2009059301A publication Critical patent/JP2009059301A/en
Withdrawn legal-status Critical Current

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently carry out a Monte Carlo evaluation by ensuring the required number of trials from the viewpoint of both reducing a calculation load and securing estimation accuracy. <P>SOLUTION: This method for determining automatic termination of the Monte Carlo evaluation goes through steps of comparing a request value Preq with failure probability Pf at that time after the termination of the i-th Monte Carlo evaluation; calculating a confidence interval upper limit value when Pf is smaller than Preq, determining that the population satisfies the request if the upper limit value is Preq or less to terminate the Monte Carlo evaluation, calculating a confidence interval lower limit value when Pf is Preq or more after the i-th evaluation, determining that the population cannot satisfy the request when the lower limit value is larger than Preq to terminate the Monte Carlo evaluation; increasing the number of trials by one excepting the above cases to carry out the (i+1)th evaluation and repeating the above procedure; and setting the actually possible maximum value Nmax of the number of trials beforehand and forcibly terminating the Monte Carlo evaluation when i exceeds Nmax. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

本発明は、様々な不確定性を乱数により発生させ、多数回のシミュレーションを実行してその統計量を得る方法であるモンテカルロ法を用い、不確定性を持つシステムを事前評価するための数値シミュレーション技術に関し、特に航空宇宙機・船舶・各種プラント等のシステム評価に適する手法である。   The present invention uses a Monte Carlo method, which is a method of generating various uncertainties by random numbers, and executing a number of simulations to obtain statistics thereof, and a numerical simulation for pre-evaluating a system having uncertainties Regarding technology, it is a method particularly suitable for system evaluation of aerospace vehicles, ships, various plants, etc.

モンテカルロ法は、乱数を用いて多数回のシミュレーションを繰り返してシステムを評価し、物理現象を統計的に推定するための数値実験である。多数の不確定要因を考慮した非線形システムの評価が可能であり、評価法として優れた性質を持つ。
モンテカルロ法は標本調査を数値的に実行するものである。真値(母集団の値)を高い精度で推定するためにはより多くのシミュレーション回数が必要となる。ところが、モンテカルロ評価のデメリットは計算負荷が高く、結果を得るまでに多くの時間を必要とすることである。よって、推定精度を高めるためにシミュレーション回数を増やせば、さらに計算時間が必要となってしまう。したがって、必要な推定精度が得られ、かつ、できるだけ少ない試行回数でモンテカルロ評価を行うことが望ましい。しかし、推定精度は評価結果に依存するため、モンテカルロ評価の実行前には推定精度は不明である。モンテカルロ評価の実行後に始めて、ミッション達成確率などの評価結果とその推定精度が算出できる。したがって、事前に必要最低限の試行回数の決定は不可能であり、実際には現実的に可能な、ある程度大きな試行回数を設定してモンテカルロ評価を実行している。
The Monte Carlo method is a numerical experiment for statistically estimating a physical phenomenon by evaluating a system by repeating a large number of simulations using random numbers. It is possible to evaluate nonlinear systems considering many uncertain factors, and it has excellent properties as an evaluation method.
The Monte Carlo method performs a sample survey numerically. More simulation times are required to estimate the true value (population value) with high accuracy. However, the disadvantage of Monte Carlo evaluation is that the calculation load is high, and it takes a lot of time to obtain the result. Therefore, if the number of simulations is increased to increase the estimation accuracy, more calculation time is required. Therefore, it is desirable to perform the Monte Carlo evaluation with as few trials as possible while obtaining the necessary estimation accuracy. However, since the estimation accuracy depends on the evaluation result, the estimation accuracy is unknown before execution of Monte Carlo evaluation. Only after the execution of the Monte Carlo evaluation, the evaluation results such as the mission achievement probability and the estimated accuracy can be calculated. Therefore, it is impossible to determine the minimum necessary number of trials in advance, and the Monte Carlo evaluation is executed by setting a somewhat large number of trials that is actually possible.

非特許文献1〜3は、それぞれ自動着陸実験、遷音速飛行実験、月着陸船の技術検討に関してモンテカルロ評価が適用されているが、いずれの場合もあらかじめシミュレーション回数を事前に設定して解析している。
また直接本発明とは異なるが、モンテカルロ評価結果の分析に関しての文献として特許文献1がある。内容は、モンテカルロ評価の結果、要求を満足できなかったケースについて、その原因となる不確定パラメータを検出する方法を提案したものである。
不確定パラメータはモンテカルロ評価の入力である。多数の不確定パラメータを同時に、かつ、それらの大きさをランダムに加えるため、評価結果が良くない場合にその原因となる不確定パラメータを見つけることは必ずしも容易ではない。特許文献1と非特許文献4〜6は、この問題について検討し、問題となる不確定パラメータの検出法を提案したものである。
特開2007−213172号公報 「モンテカルロ評価における危険不確定パラメータの検出方法」 平成19年8月23日公開 Motoda,T., Miyazawa,Y., Ishikawa,K. and Izumi,T., "Automatic Landing Flight Experiment Flight Simulation Analysis and Flight Testing", Journal of Spacecraft and Rockets, VoL36, No.4, 1999, PP.554-560 二宮哲次郎, 鈴木広一, 塚本太郎,「高速飛行実証機フェーズ2の誘導制御系評価」, JAXA-RR-04-002, 2004. Yoshiro Hamada, Tetsujiro Ninomiya, Yasuhiro Katayama, Yasuo Shinomiya, Kohtaro Matsumoto, Masayuki Yamamoto, Shunjiro Sawai, Seiya Ueno and Kentaro Hayashi, "Feasibility Study for Precise Lunar Landing using SELENE-B Lander Configuration", JAXA-RR-05-0l3E, 2005. 元田敏和,「モンテカルロ評価における危険不確定パラメータの検出法」,日本航空宇宙学会論文集, Vol.55, No.638, PP.117-124, 2007. Motoda,T. and Miyazawa Y.,"Identification of Influential Uncertaintiesin Monte Carlo Analysis", Journal of Spacecraft and Rockets, Vol39, No.4, 2002,PP.615-623. Motoda,T., "Simplified Approach to Identifying Influential Uncertainties in Monte Carlo Analysis", Journal of Spacecraft and Rockets, Vol41, No.6, 2004, PP.1071-1075.
In Non-Patent Documents 1 to 3, Monte Carlo evaluation is applied to automatic landing experiments, transonic flight experiments, and technical studies of lunar landing ships, respectively. Yes.
Although directly different from the present invention, there is Patent Document 1 as a document relating to the analysis of the Monte Carlo evaluation result. The content proposes a method for detecting uncertain parameters that cause the case where the request cannot be satisfied as a result of the Monte Carlo evaluation.
The uncertain parameter is the input for Monte Carlo evaluation. Since a large number of uncertain parameters are added at the same time and their sizes are randomly added, it is not always easy to find an uncertain parameter that causes the evaluation result when the evaluation result is not good. Patent Document 1 and Non-Patent Documents 4 to 6 examine this problem and propose a method for detecting an indeterminate parameter in question.
JP 2007-213172 “Dangerous Uncertain Parameter Detection Method in Monte Carlo Evaluation” Published on August 23, 2007 Motoda, T., Miyazawa, Y., Ishikawa, K. And Izumi, T., "Automatic Landing Flight Experiment Flight Simulation Analysis and Flight Testing", Journal of Spacecraft and Rockets, VoL36, No.4, 1999, PP.554 -560 Tetsujiro Ninomiya, Koichi Suzuki, Taro Tsukamoto, "Evaluation of Guidance Control System for High-Speed Flight Demonstration Phase 2", JAXA-RR-04-002, 2004. Yoshiro Hamada, Tetsujiro Ninomiya, Yasuhiro Katayama, Yasuhiro Shinomiya, Kohtaro Matsumoto, Masayuki Yamamoto, Shunjiro Sawai, Seiya Ueno and Kentaro Hayashi, "Feasibility Study for Precise Lunar Landing using SELENE-B Lander Configuration", JAXA-RR-05-0l3E, 2005. Toshikazu Motoda, “Detection method of uncertain risk parameters in Monte Carlo evaluation”, Journal of Japan Aerospace Society, Vol.55, No.638, PP.117-124, 2007. Motoda, T. and Miyazawa Y., "Identification of Influential Uncertainties in Monte Carlo Analysis", Journal of Spacecraft and Rockets, Vol39, No.4, 2002, PP.615-623. Motoda, T., "Simplified Approach to Identifying Influential Uncertainties in Monte Carlo Analysis", Journal of Spacecraft and Rockets, Vol41, No.6, 2004, PP.1071-1075.

上記のように、モンテカルロ評価の問題は計算負荷が大きいことであるが、それを軽減するための最低限の試行回数は、事前には設定困難である。このため実際には推定精度確保のために大きめの試行回数を設定するが、そうすると不要な計算を実行して作業効率が悪化する可能性が高くなる。また推定精度は評価結果にも依存するため、事前に設定した試行回数では推定精度が不足する可能性もある。以上の問題を勘案し、本発明が解決すべき課題は、「計算負荷軽減」と「推定精度確保」の双方の観点から、必要な試行回数を確保しつつ、効率的にモンテカルロ評価を実行することである。   As described above, the problem of Monte Carlo evaluation is that the calculation load is large, but it is difficult to set the minimum number of trials in order to reduce it. For this reason, a larger number of trials is actually set in order to ensure estimation accuracy. However, if this is done, there is a high possibility that work efficiency will deteriorate due to execution of unnecessary calculations. Further, since the estimation accuracy depends on the evaluation result, there is a possibility that the estimation accuracy is insufficient with the number of trials set in advance. Considering the above problems, the problem to be solved by the present invention is to efficiently perform Monte Carlo evaluation while securing the necessary number of trials from the viewpoints of both “reduction of calculation load” and “estimation accuracy” That is.

本発明のモンテカルロ評価の自動終了判定法は、i回目のモンテカルロ評価終了後に、そのときの失敗確率P=n/iと要求値Preqを比較するステップと、PがPreqよりも小さいときには信頼区間上限値を計算し、この上限値がPreq以下ならば「母集団は要求を満足する」と判定して、モンテカルロ評価を終了し、i回目の評価後にPがPreq以上であるときには、信頼区間下限値を計算し、この下限値がPreqよりも大きいときには「母集団は要求を満足できない」と判定して、モンテカルロ評価を終了するステップと、それ以外のときは、試行回数を1つ増やし、(i+1)回目の評価を行って上記の手順を繰り返すステップと、あらかじめ現実的に可能な、試行回数の最大値Nmaxを設定しておき、iがNmaxを超えると強制的に終了させるステップとを踏むものとした。
ただし、ここでnは失敗数を表す。
また、上記のモンテカルロ評価の自動終了判定法において、i回目の試行終了後、失敗数をn、標準正規分布における下側累積確率が危険率αに等しくなるような値をZαとして、二項分布の信頼区間の近似上限値PU1及び近似下限値PL1を次式により算出して用いるものとした。

Figure 2009059301
また、上記のモンテカルロ評価の自動終了判定法において、i回目の試行終了後、失敗数をn、標準正規分布における下側累積確率が危険率αに等しくなるような値をZαとして、信頼区間下限値を補正し、次式に基づく補正した下側信頼限界P’L1を用いるものとした。
Figure 2009059301
The automatic completion determination method of the Monte Carlo evaluation of the present invention is the step of comparing the failure probability P f = n i / i at that time with the required value P req after the completion of the i-th Monte Carlo evaluation, and P f is more than P req When it is small, the confidence interval upper limit value is calculated. If the upper limit value is equal to or less than P req, it is determined that “the population satisfies the requirement”, the Monte Carlo evaluation is terminated, and P f is equal to or greater than P req after the i-th evaluation. When the lower limit value is greater than P req , it is determined that “the population cannot satisfy the request” and the Monte Carlo evaluation is terminated, and otherwise, the number of attempts is incremented by 1, (i + 1) th evaluation performed in the step of repeating the above steps, a pre realistically, may be set the maximum value N max of the number of trials, i is N max Force over It was assumed that the steps to finish were taken.
However, where n i represents the number of failures.
Further, in the above-described Monte Carlo evaluation automatic termination determination method, after the i-th trial, the number of failures is n i , and the value that makes the lower cumulative probability equal to the risk factor α in the standard normal distribution is Z α . The approximate upper limit value P U1 and the approximate lower limit value P L1 of the confidence interval of the term distribution are calculated and used according to the following equations.
Figure 2009059301
In the above-described Monte Carlo evaluation automatic termination determination method, after the i-th trial, the number of failures is n i , and the value that makes the lower cumulative probability equal to the risk factor α in the standard normal distribution is Z α . The section lower limit value was corrected, and the corrected lower confidence limit P ′ L1 based on the following equation was used.
Figure 2009059301

本発明のモンテカルロ評価の自動終了判定法は、モンテカルロ評価を利用するにあたって、従来からの問題点は計算負荷が高いことであり、計算時間を要することであった。上記の手法を採用した本発明により不要なシミュレーションを実行しなくて済むようになるため、事前に試行回数を設定する従来のやり方に比べ、計算負荷の軽減が期待でき、効率的な解析が可能となる。
一方で、必要な推定精度確保も考慮されているため、最大許容回数Nmaxを超えない限り、試行回数が不足するということもない。
In the method of automatically ending the Monte Carlo evaluation according to the present invention, when using the Monte Carlo evaluation, the conventional problem is that the calculation load is high and calculation time is required. Since the present invention adopting the above method eliminates the need for executing unnecessary simulations, the calculation load can be reduced compared to the conventional method of setting the number of trials in advance, and efficient analysis is possible. It becomes.
On the other hand, since necessary estimation accuracy is ensured, the number of trials is not short unless the maximum allowable number Nmax is exceeded.

はじめに本発明に係るモンテカルロ評価の自動終了判定法についての全体の考え方を示し、次に具体的な方法について説明する。
[全体概要]
(1)これまでのように事前に試行回数を設定するのではなく、各回のシミュレーション(試行)終了毎に逐次失敗確率(又は、ミッション達成確率)とその推定精度(信頼区間)を算出する。真値(母集団の値)が要求値を満足するか、否か、を判断できた時点で、モンテカルロ評価を終了する。つまり、モンテカルロ評価を実行しながら、その終了を自動判定する。この概念を、図1に示す。
(2)具体的には、各試行終了時に失敗確率とその信頼区間を算出する。信頼区間の上限が要求値以下ならば、真の失敗確率は要求値以下と判定可能であるし、一方で信頼区間の下限が要求値よりも大きければ真の失敗確率は要求値以上と判定できる。この考え方を図2に示す。
(3)ところが、二項分布である失敗確率の信頼区間を解析的に求めることは不可能であり、専用のソフトや数値最適化計算を必要とするため、使い勝手が良くない。計算時間を要したり、使用するソフトの整合性の問題などが発生する。
(4)上記の問題を解決するため、容易な信頼区間の計算法を導入する。具体的には、正規分布近似を用いて、簡易な計算式で信頼区間を求められるようにする。
(5)通常の正規分布の近似式では近似誤差が大きいため、より誤差の小さい信頼区間の近似法を導出する。
(6)さらに、近似した信頼区間を用いても要求される精度を確保するため、正規分布近似の式に補正を施す。
First, the overall concept of the Monte Carlo evaluation automatic termination determination method according to the present invention will be described, and then a specific method will be described.
[Overview]
(1) Instead of setting the number of trials in advance as in the past, the failure probability (or mission achievement probability) and its estimation accuracy (confidence interval) are calculated at the end of each simulation (trial). When it is determined whether the true value (population value) satisfies the required value, the Monte Carlo evaluation is terminated. That is, the end is automatically determined while performing the Monte Carlo evaluation. This concept is illustrated in FIG.
(2) Specifically, a failure probability and its confidence interval are calculated at the end of each trial. If the upper limit of the confidence interval is less than or equal to the required value, the true failure probability can be determined to be less than the required value. On the other hand, if the lower limit of the confidence interval is greater than the required value, the true failure probability can be determined to be greater than or equal to the required value. . This concept is shown in FIG.
(3) However, it is impossible to analytically obtain a confidence interval of failure probability that is a binomial distribution, and it requires special software and numerical optimization calculation, so it is not easy to use. Calculation time is required, and there is a problem of consistency of software used.
(4) In order to solve the above problem, an easy method for calculating the confidence interval is introduced. Specifically, the confidence interval can be obtained by a simple calculation formula using normal distribution approximation.
(5) Since an approximation error is large in a normal normal distribution approximation formula, an approximation method of a confidence interval with a smaller error is derived.
(6) Further, in order to ensure the required accuracy even when the approximate confidence interval is used, the normal distribution approximation formula is corrected.

以下に、これらの詳細を記す。
[モンテカルロ評価]
図3にモンテカルロ評価法を示す。
(1)システムに存在すると考えられる不確定パラメータを設定する。不確定パラメータとは、センサ計測誤差、環境条件、初期状態、システム特性など、システムの動作に影響するが、その値が未知であるか又はその時々で値が異なるパラメータである。
(2)不確定パラメータは多数存在するが、個々のパラメータの分布を設定し、乱数を用いて各不確定パラメータを発生させる。
(3)発生させた不確定パラメータ・ベクトルがε、であり、これを組み込んだシステム・モデルを用いて、数値シミュレーションを1回実行する。
(4)結果が要求を満足していれば成功、そうでなければ失敗として評価する。
(5)乱数を用いて不確定パラメータ・ベクトルεを更新し、上記(3)−(4)の操作を多数回(N回)繰り返す。
(6)失敗ケースがn回現れた場合、失敗確率の推定値はP=n/Nとなる。
These details are described below.
[Monte Carlo evaluation]
FIG. 3 shows the Monte Carlo evaluation method.
(1) Set uncertain parameters that are considered to exist in the system. An uncertain parameter is a parameter that affects the operation of the system, such as sensor measurement error, environmental conditions, initial state, system characteristics, etc., but the value is unknown or sometimes different.
(2) Although there are many uncertain parameters, the distribution of each parameter is set and each uncertain parameter is generated using a random number.
(3) The generated uncertain parameter vector is ε i , and a numerical simulation is executed once by using a system model incorporating this.
(4) If the result satisfies the request, it is evaluated as successful, and otherwise it is evaluated as a failure.
(5) The uncertain parameter vector ε is updated using a random number, and the operations (3) to (4) are repeated many times (N times).
(6) If failure cases appear n times, the estimated failure probability is P f = n / N.

[区間推定]
モンテカルロ評価で得られる失敗確率Pは、母集団から取り出したN個の標本の値であり、必ず推定誤差を含む。そこで母集団の(真の)失敗確率を推定する必要があるが、この概念を図4に示す。真値の推定は、ある区間を指定してその中に真値が含まれる確率が[(1−α):信頼度]であるような、区間の推定を行う。αは危険率と呼ばれ、真値が区間に含まれない確率を表す。以下に、失敗確率の区間推定について記す。
(1)信頼区間の上限値をPとすれば、母集団の(真の)失敗確率がPであるとき、無作為抽出したN回の試行の失敗確率がP以下、つまり失敗数がn以下となる確率がαとなればよい。この関係を図5に示す。
(2)二項分布:試行の結果は失敗か成功の2通りしかないため、N回の試行でj回失敗する確率B(j)は二項分布となり、次式で表される。

Figure 2009059301
(3)上側信頼限界:j≦nとなる確率がαとなればよいので、Pは次式から得られる。
Figure 2009059301
(4)下側信頼限界:同様に下側信頼限界をPとすれば、αとの関係は図6で表される。このときはj≧nとなる確率がαとなればよいので、Pは次式から得られる。
Figure 2009059301
(5)上記[2],[3]式より信頼区間[P, P]が求められる。 [Section estimation]
The failure probability P f obtained by the Monte Carlo evaluation is a value of N samples extracted from the population, and always includes an estimation error. Therefore, it is necessary to estimate the (true) failure probability of the population. This concept is shown in FIG. The estimation of the true value is performed by estimating a section in which a certain section is designated and the probability that the true value is included in the section is [(1-α): reliability]. α is called a risk factor and represents the probability that the true value is not included in the interval. The following describes the interval estimation of the failure probability.
(1) If the upper limit of the confidence interval is P U, and the population's (true) failure probability is P U , the failure probability of N trials extracted at random is less than P f , that is, the number of failures The probability that is less than or equal to n may be α. This relationship is shown in FIG.
(2) Binomial distribution: Since there are only two trial results: failure or success, the probability B U (j) of failing j times in N trials is a binomial distribution and is expressed by the following equation.
Figure 2009059301
(3) the upper confidence limit: the probability that the j ≦ n may if the alpha, P U is obtained from the following equation.
Figure 2009059301
(4) Lower confidence limit: Similarly, if the lower confidence limit is P L , the relationship with α is expressed in FIG. Since the probability of j ≧ n this case may if the alpha, P L is obtained from the following equation.
Figure 2009059301
(5) The confidence interval [P L , P U ] is obtained from the above equations [2] and [3].

[区間推定の問題点]
(1) [2],[3]式により、N,n,αが与えられたときにP又はPを求める必要があるが、これらの式は解析的に解けない。このため、数値最適化計算や統計解析用のソフトウェアを必要とする。
(2) モンテカルロ評価を実行しながら、これらの式を解く場合には、N及びnを更新しながら各試行の終了毎に、多数回の計算が必要となる。
(3) このため、計算時間が増大する。または、モンテカルロ評価と区間推定計算のソフトウェアの整合性の問題が発生することもある。
[Problems of interval estimation]
(1) [2], by [3] where, N, n, alpha although it is necessary to obtain P U or P L when given, these equations can not be solved analytically. For this reason, software for numerical optimization calculation and statistical analysis is required.
(2) When solving these equations while executing Monte Carlo evaluation, a large number of calculations are required at the end of each trial while updating N and n.
(3) For this reason, calculation time increases. Or, there may be a problem of software consistency between Monte Carlo evaluation and interval estimation calculation.

[近似信頼区間の導出]
(1) 上記[2],[3]式による計算上の問題を解決するために、正規分布を利用した近似式を用いて信頼区間を求める。
(2) 通常の近似式:近似により得られる信頼区間を[PL0,PU0]とすると、一般のテキストに記述されている通常の近似式は次式で表される。

Figure 2009059301
ここで、Zαは(平均0,分散1)の標準正規分布において下側累積確率がαとなるときの、平均値(=0)からのずれを表す。この定義と数値例を図7に示す。
(3)ところが、[4]式の近似式は近似誤差が大きい。様々な試行回数Nについて、上側及び下側の信頼区間幅と失敗確率の関係を比較したものを図8に示す。図は縦軸・横軸共に対数スケールで示してある。この結果から明らかなように、[4]式の近似式では失敗確率Pが小さくなるほど、近似誤差が目立ってくる。通常は失敗確率を小さくすべくシステムの設計改善を行うものであり、特に失敗確率が小さい部分での近似誤差が小さいほうが望ましい。
(4)そこで、[4]式よりも推定誤差が小さな、より高精度な近似法を以下のように導出する。
(5)上側信頼区問について考えれば、正確な信頼区間は図5、及び[2]式によって得られる。図5において二項分布の式B(j)を正規分布で近似することになるが、この二項分布の平均mと分散σは次式となる。
Figure 2009059301
近似関数である正規分布の平均と分散もこれと等しいものとする。
(6)図9に離散分布である二項分布と連続分布である正規分布を重ねたイメージを示す。二項分布は離散分布であるので、[2]式のように下側累積確率は失敗数0→n回に対応する確率の和で表される。これに対して正規分布の累積確率は連続分布なので積分で表される。
(7)ここで、失敗数nのときの二項分布の失敗確率B(n)について考える。二項分布の場合は図の縦軸が確率B(n)そのものを表すが、正規分布の場合は確率密度関数の面積が確率を表す。よってB(n)に対応する正規分布の確率密度関数の積分区間は、失敗数n±1/2に対応する区間と考えるのが妥当である。横軸は失敗確率なので、積分区間は[(n−1/2)/N,(n+1/2)/N]となる。
(8)各失敗数j(j=0,…,n)についても同様であるため、二項分布において失敗確率Pまでの下側累積確率を求める積分区間は、正規分布の場合には図9に示すようにP’=(n+1/2)/Nまでに補正する。
(9)以上より、次式が成り立つ。
Figure 2009059301
この式から上限値Pを求めることができる。[6]式を展開してPについて整理すれば、
Figure 2009059301
これを解いて信頼区間の上限値を求めるが、この式から得られる上限値は近似値でありPとは若干異なるので、PU1と表記する。
Figure 2009059301
ここで[7-1]式からは2つの解が得られるが、これらは信頼区間の上限値及び下限値にそれぞれ対応するものである。ここでは上限値を求めることを前提に補正値P’を求めているため、大きい方の解を採用して、[8]式をPU1の値とする。
(10)信頼区間の下限値についても上記と同様の考え方で求める。今度は図6におけるPをP’=(n−1/2)/Nに補正して、次式が成り立つ。
Figure 2009059301
これを展開して整理すると、
Figure 2009059301
近似式から求める下限値なので、[8]式と同様にこれをPL1と表記すると、
Figure 2009059301
(11) [8],[11]式の近似式を用いた場合の近似誤差を図10に示す。これは図8に対応するものであり、通常の正規分布近似を用いるよりもかなり正確な値に近いことがわかる。[8],[11]式から信頼区間[PL1,PU1]は容易に計算でき、[2],[3]式を使用するときの問題は発生しない。 [Derivation of approximate confidence intervals]
(1) In order to solve the calculation problem by the above equations [2] and [3], a confidence interval is obtained using an approximate expression using a normal distribution.
(2) Normal approximate expression: When the confidence interval obtained by approximation is [P L0 , P U0 ], the normal approximate expression described in general text is expressed by the following expression.
Figure 2009059301
Here, Zα represents a deviation from the average value (= 0) when the lower cumulative probability is α in the standard normal distribution of (average 0, variance 1). FIG. 7 shows this definition and numerical examples.
(3) However, the approximation formula [4] has a large approximation error. FIG. 8 shows a comparison of the relationship between the upper and lower confidence interval widths and the failure probabilities for various trial counts N. The figure shows the logarithmic scale on both the vertical and horizontal axes. As is apparent from this result, in the approximate expression [4], the approximate error becomes more conspicuous as the failure probability P f decreases. Normally, the system design is improved to reduce the failure probability, and it is desirable that the approximation error is particularly small in a portion where the failure probability is small.
(4) Therefore, a higher-accuracy approximation method with a smaller estimation error than equation [4] is derived as follows.
(5) Considering the upper confidence zone, the exact confidence zone can be obtained from FIG. 5 and equation [2]. In FIG. 5, the binomial distribution formula B U (j) is approximated by a normal distribution. The mean m and the variance σ 2 of the binomial distribution are as follows.
Figure 2009059301
The mean and variance of the normal distribution, which is an approximate function, are also assumed to be equal to this.
(6) FIG. 9 shows an image in which a binomial distribution that is a discrete distribution and a normal distribution that is a continuous distribution are superimposed. Since the binomial distribution is a discrete distribution, the lower cumulative probability is expressed as the sum of probabilities corresponding to the number of failures 0 → n as shown in Equation [2]. On the other hand, the cumulative probability of the normal distribution is expressed as an integral because it is a continuous distribution.
(7) Now consider the failure probability B U (n) of the binomial distribution when the number of failures is n. In the case of binomial distribution, the vertical axis of the figure represents the probability B U (n) itself, but in the case of normal distribution, the area of the probability density function represents the probability. Therefore, it is appropriate to consider the integration interval of the probability density function of the normal distribution corresponding to B U (n) as the interval corresponding to the number of failures n ± 1/2. Since the horizontal axis is the failure probability, the integration interval is [(n-1 / 2) / N, (n + 1/2) / N].
(8) Since the same applies to the number of failures j (j = 0,..., N), the integration interval for obtaining the lower cumulative probability up to the failure probability P f in the binomial distribution is shown in FIG. P 'f = as shown in 9 (n + 1/2) / n to the corrected.
(9) From the above, the following equation holds.
Figure 2009059301
The upper limit value P U can be obtained from this equation. In summary for P U Expand the [6] formula,
Figure 2009059301
By solving this, the upper limit value of the confidence interval is obtained. The upper limit value obtained from this equation is an approximate value and is slightly different from P U , so it is expressed as P U1 .
Figure 2009059301
Here, two solutions are obtained from the equation [7-1], and these correspond to the upper limit value and the lower limit value of the confidence interval, respectively. Here, since the correction value P ′ f is obtained on the assumption that the upper limit value is obtained, the larger solution is adopted and the equation [8] is set as the value of P U1 .
(10) The lower limit value of the confidence interval is determined in the same way as described above. This time, P f in FIG. 6 is corrected to P ′ f = (n−1 / 2) / N, and the following equation is established.
Figure 2009059301
If this is expanded and organized,
Figure 2009059301
Since this is the lower limit value obtained from the approximate expression, if this is expressed as P L1 as in [8],
Figure 2009059301
(11) FIG. 10 shows an approximation error in the case of using the approximate expressions [8] and [11]. This corresponds to FIG. 8, and it can be seen that it is much closer to a more accurate value than using the normal normal distribution approximation. The confidence interval [P L1 , P U1 ] can be easily calculated from the equations [8] and [11], and no problem occurs when the equations [2] and [3] are used.

[近似信頼区間の補正]
(1) [8],[11]式の近似式を用いた場合、下側信頼区間幅はやや小さめ、上側信頼区間幅はやや大きめになる(図10)。
(2) モンテカルロ評価の終了判定では、図2に示すように要求値が信頼区間の外側にあるとき終了判定する。近似による信頼区間幅が正確な値よりも大きい場合に要求値がその外側にあれば、その要求値は必ず正確な信頼区間の外側となる。従って、この場合は近似で評価しても正確な信頼区間で評価した場合と結果は変わらない。
(3) ところが、近似による信頼区間幅が正確ものよりも小さい場合には、近似では終了判定できても、要求値は正確な信頼区間の外側であるとは限らない。この場合は終了判定が拙速となる可能性がある。
(4) 図10に示すように[11]式から得られる下限値を用いると、下側の信頼区間幅は正確な値よりも小さくなる。このまま用いると終了判定が拙速となるため、下側の信頼区間幅が正確な値よりも大きくなるように補正を行う。
(5) [11]式の近似による信頼区間幅をΔPL1、補正した下側信頼限界をP’L1として、下側信頼区間幅を広げるため次のようにおく。

Figure 2009059301
ここで、λは1より大きい定数であり、以下でλの値について検討する。
(6) P’L1が正確な信頼限界Pよりも小さくなればよいので、
Figure 2009059301
よって、λは正確な信頼区間幅ΔPと近似区間幅ΔPL1の比よりも大きくなればよい。
(7) 様々な試行回数N、及びαについてΔP/ΔPL1を計算した結果を図11に示す。この結果から、実用的なN,αの範囲においては、λは1.2以下であることがわかる。したがって、λ=1.2として[12]式から下限値を求めれば、終了判定が拙速となることはない。
(8) 以上より、補正を加えた信頼区間の下側限界値を、次式により得る。
Figure 2009059301
[Correction of approximate confidence interval]
(1) When the approximate expressions [8] and [11] are used, the lower confidence interval width is slightly smaller and the upper confidence interval width is slightly larger (FIG. 10).
(2) In the end determination of the Monte Carlo evaluation, the end determination is made when the required value is outside the confidence interval as shown in FIG. If the confidence interval width by approximation is larger than an accurate value, if the required value is outside the accurate value, the required value is always outside the accurate confidence interval. Therefore, in this case, even if the evaluation is performed by approximation, the result is the same as the case where the evaluation is performed with an accurate confidence interval.
(3) However, if the confidence interval width by approximation is smaller than the accurate one, the required value is not always outside the accurate confidence interval even if the approximation can be determined to end. In this case, the end determination may be slow.
(4) As shown in FIG. 10, when the lower limit value obtained from the equation [11] is used, the lower confidence interval width becomes smaller than an accurate value. If it is used as it is, the end determination becomes a rapid speed, so that the lower confidence interval width is corrected to be larger than an accurate value.
(5) Assuming that the confidence interval width according to the approximation of the equation [11] is ΔP L1 and the corrected lower confidence limit is P ′ L1 , the following is made to widen the lower confidence interval width.
Figure 2009059301
Here, λ is a constant larger than 1, and the value of λ will be considered below.
(6) Since P ′ L1 needs to be smaller than the accurate confidence limit P L ,
Figure 2009059301
Therefore, it is sufficient that λ is larger than the ratio between the accurate confidence interval width ΔP L and the approximate interval width ΔP L1 .
(7) FIG. 11 shows the results of calculating ΔP L / ΔP L1 for various trials N and α. From this result, it is understood that λ is 1.2 or less in a practical range of N and α. Therefore, if λ = 1.2 and the lower limit value is obtained from the equation [12], the end determination will not be slow.
(8) From the above, the lower limit value of the confidence interval with correction is obtained by the following equation.
Figure 2009059301

[自動終了判定手順]
(1) 終了判定手順の概略は図1に示したが、ここでは上記の内容を踏まえ、具体的な手順について記す。自動終了判定アルゴリズムを図12に示す。
(2) i回目のモンテカルロ評価終了後に、そのときの失敗確率P=n/iと要求値Preqを比較する。PがPreqよりも小さいときには[8]式よりPU1を計算し、このPU1がPreq以下ならば「母集団は要求を満足する」と判定して、モンテカルロ評価を終了する。
(3) 一方、i回目の評価後にPがPreq以上であるときには、[14]式よりP’L1を計算し、このP’L1がPreqよりも大きいときには「母集団は要求を満足できない」と判定して、モンテカルロ評価を終了する。
(4) 上記のように判定できないときには、試行回数を1つ増やし、(i+1)回目の評価を行って(2)〜(3)の手順を繰り返す。
(5) ここで、試行回数を増やしたときにPと要求値Preqの値がかなり近い場合には、終了判定するためにはかなり小さな信頼区間幅が必要となる。小さな信頼区間幅を得るためには、図10からも明らかなように試行回数を増やさざるを得ない。上記の終了判定アルゴリズムに従って計算を続けると、試行回数が非現実的な値にまで増加する可能性もある。
(6) これを防ぐため、評価時間や計算負荷を考慮してあらかじめ許容可能な最大試行回数Nmaxを定めておく。試行回数がNmaxに達したら、終了判定できなくても強制的にモンテカルロ評価を終了させ、その時点の計算結果を保存する。
以上の手順により、モンテカルロ評価の自動終了判定を行う。
[Automatic termination judgment procedure]
(1) Although the outline of the end determination procedure is shown in FIG. 1, here, a specific procedure is described based on the above contents. The automatic termination determination algorithm is shown in FIG.
(2) After completion of the i-th Monte Carlo evaluation, the failure probability P f = n i / i at that time is compared with the required value P req . When P f is smaller than P req, P U1 is calculated from equation [8]. If this P U1 is equal to or less than P req, it is determined that “the population satisfies the requirement”, and the Monte Carlo evaluation is terminated.
(3) On the other hand, when P f is greater than or equal to P req after the i-th evaluation, P ′ L1 is calculated from the equation [14]. When P ′ L1 is greater than P req , “the population satisfies the requirement” Monte Carlo evaluation is terminated.
(4) If the determination cannot be made as described above, the number of trials is increased by one, the (i + 1) th evaluation is performed, and the procedures (2) to (3) are repeated.
(5) Here, if the number of trials is increased and the value of P f and the required value P req are quite close, a considerably small confidence interval width is required to determine the end. In order to obtain a small confidence interval, the number of trials must be increased, as is apparent from FIG. If the calculation is continued according to the above end determination algorithm, the number of trials may increase to an unrealistic value.
(6) In order to prevent this, an allowable maximum number of trials N max is determined in advance in consideration of evaluation time and calculation load. When the number of trials reaches N max , the Monte Carlo evaluation is forcibly terminated even if the termination determination cannot be made, and the calculation result at that time is stored.
The automatic termination determination of the Monte Carlo evaluation is performed by the above procedure.

ここで述べたモンテカルロ評価の自動終了判定法を、自動着陸実験のモンテカルロ評価に適用した例を示す。
[自動着陸実験]
自動着陸実験(ALFLEX)は、宇宙往還機の自動着陸技術を確立するために1996年に実施されたものである。実験機はヘリコプタに吊されて飛行し、その後滑走路手前2700m、高度1500mにおいて分離される。その後実験機は各種センサ信号を搭載計算機に取り込み、誘導制御ロジックに従ってコントロールされ、自動飛行により滑走路に着陸する。実験の概要を図13に示す。
An example in which the Monte Carlo evaluation automatic end determination method described here is applied to Monte Carlo evaluation of an automatic landing experiment will be described.
[Automatic landing experiment]
The Automatic Landing Experiment (ALFLEX) was conducted in 1996 to establish automatic landing technology for spacecraft. The test aircraft flies in a helicopter, and then separated at 2700m before the runway and at an altitude of 1500m. After that, the experimental aircraft captures various sensor signals into the onboard computer, is controlled according to the guidance control logic, and lands on the runway by automatic flight. An outline of the experiment is shown in FIG.

[モンテカルロ評価]
本実験では自動着陸実験前に試験飛行による動作確認を実施することは困難であり、初飛行時に自動着陸を成功させなければならない。従って、システムの事前評価が非常に重要となる。事前評価のためにシステムの数学モデルを構築するが、現実にはセンサ計測誤差、空気力の誤差、質量や慣性モーメントの誤差、風条件、初期条件等の様々な不確定要因が存在し、数学モデルでは記述できない要素が多く存在する。モンテカルロ評価は、これらの不確定パラメタの影響も含めた評価が可能な手段であり、本実験の事前評価手法として最適である。
本実験のモンテカルロ評価において加えた不確定パラメータの数は約100であり、その項目の概要を図14に示す。モンテカルロ評価では、これらの各不確定パラメータの分布を仮定し、その分布に基づき乱数を用いて不確定パラメータの値を発生させる。これらを同時に加え、多数回のシミュレーションを実行することによってミッション達成確率を推定する。
[Monte Carlo evaluation]
In this experiment, it is difficult to confirm the operation by test flight before the automatic landing test, and automatic landing must be successful at the first flight. Therefore, prior evaluation of the system is very important. Although a mathematical model of the system is built for prior evaluation, in reality there are various uncertain factors such as sensor measurement error, aerodynamic error, mass and inertial error, wind condition, initial condition, etc. There are many elements that cannot be described in the model. Monte Carlo evaluation is a means that enables evaluation including the effects of these uncertain parameters, and is the most suitable prior evaluation method for this experiment.
The number of uncertain parameters added in the Monte Carlo evaluation of this experiment is about 100, and an overview of the items is shown in FIG. In the Monte Carlo evaluation, a distribution of each of these uncertain parameters is assumed, and a value of the uncertain parameter is generated using a random number based on the distribution. By adding these simultaneously, the mission achievement probability is estimated by executing many simulations.

[評価基準]
ミッション達成(成功)、または不達成(失敗)は、着陸する際の接地時パラメータで規定される。表1に接地時に評価されるパラメータと、その許容範囲を示す。これらのパラメータが全て許容範囲内であれば成功、一つでも外れていれば失敗とカウントして、失敗確率を算出する。

Figure 2009059301
[Evaluation criteria]
Mission completion (success) or non-achievement (failure) is defined by the ground contact parameters when landing. Table 1 shows the parameters evaluated at the time of grounding and their allowable ranges. If all of these parameters are within the allowable range, the failure probability is calculated by counting the success, and if even one of them is out of the range, the failure is counted.
Figure 2009059301

[自動終了判定結果]
失敗確率の要求値P=0.Olと設定して、図12に示す自動終了判定手順に従って、モンテカルロ評価を実施した。つまり、この場合は母集団の失敗確率が1%以下であるか、又は1%を超えてしまうかが確認できた時点で評価を自動的に終了する。ここで、信頼区間算出のための危険率はα=0.05と設定した。
このときの自動判定結果を表2に示す。試行回数は512回を要しており、失敗ケースが11回であった。この標本の失敗確率Pは0.0215であり要求値を超えていることがわかる。このとき、自動終了判定で使用した信頼区間の下限値がP’L1=0、0106でありPreqよりも大きいため、母集団は失敗確率の要求Preqを満足することができないと判定している。
確認のために、最終的な結果について正確な下限値を求めてみるとP=0.0121であり、正確な信頼区間を用いてもやはり要求Preqを満足することができないという結果になることが確認できる。

Figure 2009059301
[Automatic end judgment result]
Monte Carlo evaluation was performed according to the automatic termination determination procedure shown in FIG. 12 with the failure probability request value P = 0.Ol. That is, in this case, the evaluation is automatically terminated when it is confirmed whether the failure probability of the population is 1% or less or exceeds 1%. Here, the risk factor for calculating the confidence interval was set to α = 0.05.
Table 2 shows the automatic determination result at this time. The number of trials required 512 times, and 11 failed cases. It can be seen that the failure probability P f of this sample is 0.0215, which exceeds the required value. At this time, since the lower limit of the confidence interval used in automatic termination determination is greater than is P req is P 'L1 = 0,0106, it is determined that the population can not satisfy the request P req failure probability Yes.
For confirmation, when an accurate lower limit is obtained for the final result, P L = 0.0121, and even if an accurate confidence interval is used, the request P req cannot be satisfied. I can confirm.
Figure 2009059301

次に、失敗確率の要求値Preq=0.05とした場合の結果を表3に示す。今度は試行回数189回、失敗数4回となり、標本の失敗確率はP=0.0212となった。このとき、終了判定に用いた信頼区間上限値はPU1=0.0498であり、これは要求値Preqよりも小さい。よって、この場合は失敗確率の要求Preqを満足するという結果となった。二項分布の正確な上限値を計算してみるとP=0.0478であり、正確な信頼区間を用いても要求を満足するという結果であることが確認できる。
また、表1と比較すると評価回数がかなり少ない結果となっているが、これは要求値Preqと失敗確率Pがある程度離れているために、判定に必要な信頼区間幅はそれほど小さくなくてもよかったためである。
従来は評価回数を事前に設定してモンテカルロ評価を実施していたが、自動終了判定法を用いれば、このように少ない試行回数で済むこともあり、無駄な評価を省略でき、効率的に作業を進めることが可能となる。

Figure 2009059301
Next, Table 3 shows the results when the failure probability request value P req = 0.05. This time, the number of trials 189 times, will fail number four times, failure probability of the sample it became P f = 0.0212. At this time, the confidence interval upper limit value used for the end determination is P U1 = 0.0498, which is smaller than the required value P req . Therefore, in this case, the result is that the failure probability request P req is satisfied. When calculating an accurate upper limit value of the binomial distribution, P U = 0.0478, and it can be confirmed that the result satisfies the requirement even if an accurate confidence interval is used.
In addition, the number of evaluations is considerably smaller than that in Table 1. This is because the required value P req and the failure probability P f are separated to some extent, so the confidence interval width necessary for determination is not so small. It was because it was good.
Conventionally, Monte Carlo evaluation was performed with the number of evaluations set in advance, but using the automatic termination judgment method can reduce the number of trials in this way, eliminating unnecessary evaluations and enabling efficient work. It is possible to proceed.
Figure 2009059301

最後に要求値Preqが標本の失敗確率に非常に近い場合の例として、Preq=0.02とした場合の自動終了判定の結果を表4に示す。この場合は、母集団が要求値を満足するか否かを判断するために、9,175回と多数の試行回数を必要としていることがわかる。このように、Pが要求値Preqにかなり近い値になる場合には、終了判定により小さな信頼区間幅が必要となるために多くの試行回数が必要となる。場合によっては、モンテカルロ評価の計算がいつまでも終了しないという状況も起こりうる。したがって、ここで述べた自動終了判定法を使用する際には、あらかじめ許容可能な最大評価回数Nmaxを設定しておき、評価回数がNmaxを超えたら判定できなくても評価計算を終了させるようにしておくのが実用的である。

Figure 2009059301
Finally, as an example of the case where the required value P req is very close to the failure probability of the sample, the result of the automatic end determination when P req = 0.02 is shown in Table 4. In this case, it can be seen that a large number of trials of 9,175 times are required to determine whether the population satisfies the required value. As described above, when P f becomes a value that is quite close to the required value P req , a small confidence interval width is required for the end determination, and thus a large number of trials are required. In some cases, a situation may occur where the calculation of Monte Carlo evaluation does not end indefinitely. Therefore, when using the automatic end determination method described here, an allowable maximum number of evaluations N max is set in advance, and the evaluation calculation is ended even if the determination cannot be made when the number of evaluations exceeds N max . It is practical to do so.
Figure 2009059301

環境条件・システムの特性・初期条件等様々な不確定パラメータの存在下で動作することが要求されるシステムに対して、事前検討としてモンテカルロ評価が有効であるシステムに利用可能である。例としては、航空宇宙機・船舶・各種プラント等の評価が有効であると考えられる。   It can be used for a system in which Monte Carlo evaluation is effective as a preliminary study for a system that is required to operate in the presence of various uncertain parameters such as environmental conditions, system characteristics, and initial conditions. As an example, it is considered effective to evaluate aerospace vehicles, ships, various plants, and the like.

モンテカルロ評価の自動終了判定法の概念図である。It is a conceptual diagram of the automatic completion | finish determination method of a Monte Carlo evaluation. モンテカルロ評価結果の判定を説明する図である。It is a figure explaining determination of a Monte Carlo evaluation result. モンテカルロ評価の概念図である。It is a conceptual diagram of a Monte Carlo evaluation. 標本調査の概念を説明する図である。It is a figure explaining the concept of sample investigation. 上側信頼限界とαの関係を示す図である。It is a figure which shows the relationship between an upper reliability limit and (alpha). 下側信頼限界とαの関係を示す図である。It is a figure which shows the relationship between a lower reliability limit and (alpha). αの定義と数値例を示す図である。It is a figure which shows the definition and numerical example of Z ( alpha ). 信頼区間幅の近似誤差を示す図である。It is a figure which shows the approximation error of a confidence interval width. 正規分布近似による累積確率を説明する図である。It is a figure explaining the accumulation probability by normal distribution approximation. 改良した近似法による信頼区間幅の近似誤差を説明する図である。It is a figure explaining the approximation error of the confidence interval width by the improved approximation method. 様々なα,Nに対するΔP/ΔPL1の値を示す図である。Various alpha, illustrates the value of ΔP L / ΔP L1 against N. 自動終了判定のアルゴリズムを説明する図である。It is a figure explaining the algorithm of automatic end determination. 航空機の自動着陸実験を説明する図である。It is a figure explaining the automatic landing experiment of an aircraft. 不確定パラメータを説明する図である。It is a figure explaining an indeterminate parameter.

Claims (3)

i回目のモンテカルロ評価終了後に、そのときの失敗確率P=n/iと要求値Preqを比較するステップと、PがPreqよりも小さいときには信頼区間上限値を計算し、この上限値がPreq以下ならば「母集団は要求を満足する」と判定して、モンテカルロ評価を終了し、i回目の評価後にPがPreq以上であるときには、信頼区間下限値を計算し、この下限値がPreqよりも大きいときには「母集団は要求を満足できない」と判定して、モンテカルロ評価を終了するステップと、それ以外のときは、試行回数を1つ増やし、(i+1)回目の評価を行って上記の手順を繰り返すステップと、あらかじめ現実的に可能な、試行回数の最大値Nmaxを設定しておき、iがNmaxを超えると強制的に終了させるステップとを踏むものであるモンテカルロ評価の自動終了判定法。
ただし、ここでnは失敗数を表す。
After completion of the i-th Monte Carlo evaluation, a step of comparing the failure probability P f = n i / i at that time with the required value P req , and when P f is smaller than P req , a confidence interval upper limit value is calculated, and this upper limit is calculated If the value is equal to or less than P req, it is determined that “the population satisfies the requirement”, and the Monte Carlo evaluation is terminated. When P f is equal to or greater than P req after the i-th evaluation, the lower limit value of the confidence interval is calculated. when this lower limit is greater than P req it is determined that the "population can not satisfy the request", a step of terminating the Monte Carlo evaluation, at other times, increasing by one the number of trials, (i + 1) A step of repeating the above procedure by performing a second evaluation, and a step of setting a maximum number N max of trials that can be realistic in advance and forcibly ending when i exceeds N max are performed. Monteca Automatic termination judgment method for Luro evaluation.
However, where n i represents the number of failures.
i回目の試行終了後、失敗数をn、標準正規分布における下側累積確率が危険率αに等しくなるような値をZαとして、二項分布の信頼区間の近似上限値PU1及び近似下限値PL1を次式により算出して用いる請求項1に記載のモンテカルロ評価の自動終了判定法。
Figure 2009059301
After completion of the i-th trial, the number of failures is n i , and the value that makes the lower cumulative probability in the standard normal distribution equal to the risk factor α is Z α and the approximate upper limit value P U1 of the confidence interval of the binomial distribution and the approximation The Monte Carlo evaluation automatic end determination method according to claim 1, wherein the lower limit value P L1 is calculated and used according to the following equation.
Figure 2009059301
i回目の試行終了後、失敗数をn、標準正規分布における下側累積確率が危険率αに等しくなるような値をZαとして、信頼区間下限値を補正し、次式に基づく補正した下側信頼限界P’L1を用いる請求項1に記載のモンテカルロ評価の自動終了判定法。
Figure 2009059301
After completion of the i-th trial, the number of failures is n i , the value that makes the lower cumulative probability in the standard normal distribution equal to the risk factor α is Z α , the confidence interval lower limit value is corrected, and correction is performed based on the following equation The automatic termination determination method for Monte Carlo evaluation according to claim 1, wherein the lower reliability limit P'L1 is used.
Figure 2009059301
JP2007228079A 2007-09-03 2007-09-03 Method for determining automatic termination of monte carlo evaluation Withdrawn JP2009059301A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2007228079A JP2009059301A (en) 2007-09-03 2007-09-03 Method for determining automatic termination of monte carlo evaluation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2007228079A JP2009059301A (en) 2007-09-03 2007-09-03 Method for determining automatic termination of monte carlo evaluation

Publications (1)

Publication Number Publication Date
JP2009059301A true JP2009059301A (en) 2009-03-19

Family

ID=40554949

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2007228079A Withdrawn JP2009059301A (en) 2007-09-03 2007-09-03 Method for determining automatic termination of monte carlo evaluation

Country Status (1)

Country Link
JP (1) JP2009059301A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012033134A (en) * 2010-03-29 2012-02-16 Bank Of Tokyo-Mitsubishi Ufj Ltd Prediction device, program and prediction method
CN110532513A (en) * 2019-07-11 2019-12-03 肇庆学院 Vehicle rollover prediction algorithm based on radius importance sampling failure probability method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012033134A (en) * 2010-03-29 2012-02-16 Bank Of Tokyo-Mitsubishi Ufj Ltd Prediction device, program and prediction method
CN110532513A (en) * 2019-07-11 2019-12-03 肇庆学院 Vehicle rollover prediction algorithm based on radius importance sampling failure probability method

Similar Documents

Publication Publication Date Title
Kraft The air force digital thread/digital twin-life cycle integration and use of computational and experimental knowledge
Paiva et al. Robust and reliability-based design optimization framework for wing design
Lombaerts et al. Safe maneuvering envelope estimation based on a physical approach
CN104317990A (en) Multi-stage task spacecraft reliability improving method based on risks
Dean et al. High resolution CFD simulations of maneuvering aircraft using the CREATE-AV/Kestrel solver
Varrier et al. A parity space-based fault detection on lpv systems: Approach for vehicle lateral dynamics control system
Grauer Real-time data-compatibility analysis using output-error parameter estimation
Abdol-Hamid et al. Ares I Vehicle Computed Turbulent Ascent Aerodynamic Data Development and Analysis
Abdol-Hamid et al. Overview of Ares-I CFD ascent aerodynamic data development and analysis based on USM3D
JP2009059301A (en) Method for determining automatic termination of monte carlo evaluation
CN112329137A (en) Carrier rocket online orbit-entering capability evaluation method based on balanced flight theory
Clifton et al. Determining the stability and control characteristics of high-performance maneuvering aircraft using high-resolution CFD simulation with and without moving control surfaces
Safie et al. Reliability and probabilistic risk assessment—How they play together
van den Brandt et al. Safe flight envelope uncertainty quantification using probabilistic reachability analysis
Cao et al. Towards assurance case evidence generation through search based testing: work-in-progress
Madson et al. F-16/generic store carriage load predictions at transonic Mach numbers using TranAir
Huang et al. Uncertainty analysis of reachable set for planetary entry using polynomial chaos
Wignall et al. Uncertainty quantification for launch vehicle aerodynamic lineloads
Halstrom et al. Comparison of OVERFLOW Computational and Experimental Results for a Blunt Mars Entry Vehicle Concept During Supersonic Retropropulsion
CN109255171B (en) Method for automatically judging convergence of numerical simulation calculation
Zaitseva et al. Reliability evaluation of heterogeneous drone fleet by structure function based method
Zhang et al. Database building and interpolation for a safe flight envelope prediction system
Motoda et al. Identification of influential uncertainties in monte carlo analysis
d'Ippolito et al. Reliability-based design optimization of composite and steel aerospace structures
Motoda Quick detection of influential inputs in Monte Carlo simulation

Legal Events

Date Code Title Description
A300 Application deemed to be withdrawn because no request for examination was validly filed

Free format text: JAPANESE INTERMEDIATE CODE: A300

Effective date: 20101207