JP2006072820A - Device for analyzing combination optimization problem - Google Patents

Device for analyzing combination optimization problem Download PDF

Info

Publication number
JP2006072820A
JP2006072820A JP2004257120A JP2004257120A JP2006072820A JP 2006072820 A JP2006072820 A JP 2006072820A JP 2004257120 A JP2004257120 A JP 2004257120A JP 2004257120 A JP2004257120 A JP 2004257120A JP 2006072820 A JP2006072820 A JP 2006072820A
Authority
JP
Japan
Prior art keywords
evaluation function
solution
partial evaluation
state
partial
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.)
Pending
Application number
JP2004257120A
Other languages
Japanese (ja)
Inventor
Nobuhiko Itaya
伸彦 板屋
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.)
Mitsubishi Electric Corp
Original Assignee
Mitsubishi Electric Corp
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 Mitsubishi Electric Corp filed Critical Mitsubishi Electric Corp
Priority to JP2004257120A priority Critical patent/JP2006072820A/en
Publication of JP2006072820A publication Critical patent/JP2006072820A/en
Pending legal-status Critical Current

Links

Images

Landscapes

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

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for analyzing a combination optimization problem for efficiently escaping from a local solution. <P>SOLUTION: This analysis device is provided with a means for starting with an initial combination status, and for deciding a transition object status by an evaluation function from combination statuses defined as adjacent statuses, and for successively and repeatedly retrieving transition for searching the optimal combination status for minimizing or maximizing the evaluation function, and for comparing the most satisfactory solution of the combination status whose evaluation function value is the most satisfactory with the evaluation function value of the current status, and for defining this as the most satisfactory solution when the current status is satisfactory, a means for, when the evaluation function value of the current status is a local solution better than the evaluation function values of all the adjacent statuses, setting partial evaluation function improvement conditions that any one partial evaluation function value is better than the local solution, a means for defining only the adjacent status satisfying the conditions among the adjacent statuses of the current status as the object of retrieval while the conditions are set, a means for releasing the above conditions for successful transition to such a status that the evaluation function is better than the local solution and a means for returning the current status to the local solution set with the above conditions, and for continuing retrieval when the current status is turned into the local solution while the conditions are set. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

この発明は、例えば1つ1つの組合せの状態の集まりである電力供給状態における様々な状態の組合せの中から最も良い組合せを評価関数を用いて解析して求める組合せ最適化問題の解析装置に関する。   The present invention relates to an analysis apparatus for a combination optimization problem that is obtained by analyzing the best combination among various combinations in a power supply state, which is a collection of states of each combination, for example, using an evaluation function.

従来、この種の組合せ最適化問題の解法として、シミュレーテッドアニーリング法(Simulated Annealing:SA)という、物理系における焼きなまし(annealing)過程を模擬した方法に基づくものがある。焼きなましとは、固体に熱を加えて溶融し、それを徐々に冷やすことにより結晶を生成するプロセスである。固体にエネルギーを加えて温度を上げ、溶融状態にすれば粒子はランダムな配置となる。その状態からゆっくりと温度を下げることにより、最も内部エネルギーの小さい基底状態となり、このときの粒子は規則正しい配置、すなわち結晶となる。SA法は、このような粒子の挙動を組合せ最適化問題の決定変数に対応させ、焼きなまし過程を計算機上で模擬するものである。   Conventionally, as a method for solving this kind of combinatorial optimization problem, there is a method based on a simulated annealing (SA) method that simulates an annealing process in a physical system. Annealing is a process of producing crystals by applying heat to a solid to melt it and gradually cooling it. When energy is applied to the solid to raise the temperature and bring it into a molten state, the particles are randomly arranged. By slowly lowering the temperature from that state, the ground state with the smallest internal energy is obtained, and the particles at this time are regularly arranged, that is, crystals. In the SA method, the behavior of particles is made to correspond to the decision variable of the combinatorial optimization problem, and the annealing process is simulated on a computer.

最小化の場合のSA法では、まず組合せ状態を表すxと温度を表すTの初期値を設定し、xの隣の状態である近接解yを生成してyとxの目的関数(評価関数)の変化量ΔEを計算し、ΔEが負(yの方がxより良い解)ならxをyに更新する(xをyに遷移させる)。またΔEが非負なら所定区間の一様乱数γを発生させ、Exp(−ΔE/T)>γが成立するか否かの判定をして(成立する確率はΔEが小さく(xに比べたyの解の悪化度合いが小さく)温度Tが大きい程大きい)、成立する場合には解の悪化を許容しyを採用し、成立しない場合にはxの更新は行わない。そして平衡状態(多数回この判定を行ってもxに変化がない状態)か否かの判定をして、平衡状態でなければ次の隣の状態である近接解yを生成して同様な処理を繰り返し、平衡状態なら温度Tを下げるためにTをρ(0<ρ<1)倍する。そして基底状態(充分に温度Tが低い状態)になったか否かの判定を行い、基底状態ならその時点のxを最適解として処理終了とし、基底状態でなければ次の隣の状態である近接解yを生成して同様な処理を繰り返えす(非特許文献1参照)。このような方法では、より悪い状態への遷移が確率的に許されており、xが極小解に到達しても、極小解に捕まらずに遷移を継続することが可能となり、より良い解を見つけ出すことができる。なお、評価関数の最小化を目的とする場合の極小解、または最大化の場合の最大解を、局所解と呼ぶ。   In the SA method in the case of minimization, first, an initial value of x representing a combined state and T representing a temperature is set, a proximity solution y that is a state adjacent to x is generated, and an objective function (evaluation function) of y and x is generated. ) And if ΔE is negative (y is a better solution than x), x is updated to y (x is changed to y). If ΔE is non-negative, a uniform random number γ in a predetermined interval is generated, and it is determined whether Exp (−ΔE / T)> γ is satisfied (the probability of being satisfied is that ΔE is small (y compared to x) The degree of deterioration of the solution is small (the larger the temperature T is, the larger the value is)). If satisfied, the deterioration of the solution is allowed and y is adopted. If not satisfied, x is not updated. Then, it is determined whether or not it is in an equilibrium state (a state in which x does not change even if this determination is made many times), and if it is not in an equilibrium state, a proximity solution y that is the next adjacent state is generated and similar processing is performed. In an equilibrium state, T is multiplied by ρ (0 <ρ <1) to lower the temperature T. Then, it is determined whether or not a ground state (a state in which the temperature T is sufficiently low) has been reached. If the ground state, the process ends with x as an optimal solution, and if it is not the ground state, the next adjacent state is the adjacent state. A solution y is generated and the same processing is repeated (see Non-Patent Document 1). In such a method, a transition to a worse state is probabilistically allowed, and even if x reaches a minimal solution, it is possible to continue the transition without being caught by the minimal solution, and a better solution can be obtained. You can find out. Note that the minimal solution for minimizing the evaluation function or the maximum solution for maximization is called a local solution.

櫻井良文他著、「新しい電力システム計画手法」電気学会技術報告、第647号、pp51〜52、1997年8月Yoshifumi Sakurai et al., “New Power System Planning Method”, IEEJ Technical Report, 647, pp 51-52, August 1997

このような従来の組合せ最適化問題の解析装置において状態の組合せ最適化問題を解く上での最大の技術課題は、いかに局所解から脱出し探索を継続するかにあり、従来から様々な方法が提案されている。これらを大きく分類すると下記となる。
(1)ある確率でより悪い状態への遷移を許す(上述の方法)。
(2)許される遷移の中で最良の状態への遷移を繰り返す(局所解においては評価関数値が悪化する遷移が発生する)こととし、遷移に伴い組合せの内容に変化のあった部分に関しては、ある期間、逆向きの変化を禁止することにより、通過した局所解へ戻ることを抑制する。
(3)許される遷移の中で最良の状態への遷移を繰り返す(局所解においては評価関数値が悪化する遷移が発生する)こととし、通過した局所解の評価関数値には以降下駄を履かせて評価を悪化させて評価することとし、通過した局所解へ戻ることを抑制する。
In such a conventional combinatorial optimization problem analyzer, the greatest technical problem in solving the state combinatorial optimization problem is how to continue escape search from a local solution. Proposed. These are broadly classified as follows.
(1) Allow a transition to a worse state with a certain probability (the method described above).
(2) The transition to the best state among the permitted transitions is repeated (in the local solution, a transition that deteriorates the evaluation function value occurs). By prohibiting reverse changes for a certain period, it is possible to suppress the return to the local solution that has passed.
(3) The transition to the best state among the allowed transitions is repeated (a transition where the evaluation function value deteriorates in the local solution occurs), and the evaluation function value of the passed local solution is subsequently clogged. The evaluation is made worse and the evaluation is made, and the return to the local solution that has passed is suppressed.

この発明は、組合せ最適化問題を解く上での最大の技術課題である「いかに局所解から脱出し探索を継続するか」に関して、より効率的な方法で探索を継続するようにした組合せ最適化問題の解析装置を提供することを目的とする。   The present invention relates to a combination optimization in which a search is continued in a more efficient manner with respect to "how to continue a search by exiting from a local solution", which is the greatest technical problem in solving a combination optimization problem. The object is to provide a problem analysis device.

この発明は、初期の組合せ状態から出発し、隣の状態と定義された組合せ状態の中から評価関数を用いて遷移すべき状態を決定し、順次この遷移先への遷移を繰り返す探索を行うことにより、評価関数を最小化または最大化する最適な組合せ状態を求めるコンピュータを用いた組合せ最適化問題の解析装置であって、前記評価関数が複数の部分評価関数値による関数として表現され、すべての部分評価値の領域においてすべての部分評価関数値が増加すると評価関数値が増加または変化しないことが保証されており、探索した中で評価関数値が最もよい組合せ状態である最良解と現在の組合せ状態である現在状態の評価関数値を比較し、現在状態の評価関数値の方がよい場合に現在状態を最良解とする最良解更新手段と、現在状態がその評価関数値がすべての隣接状態の評価関数値よりもよい局所解である場合に、局所解よりも何れか1個の部分評価関数値がよいことである部分評価関数改善条件を設定する部分評価関数改善条件設定手段と、部分評価関数改善条件が設定されている場合は、現在状態の隣接状態の中で部分評価関数改善条件を満足する隣接状態だけを探索の対象とする探索範囲限定手段と、局所解よりも評価関数がよい状態に遷移できた際に部分評価関数改善条件を解除する部分評価関数改善条件解除手段と、部分評価関数改善条件が設定されている場合に現在状態が局所解になった場合に、部分評価関数改善条件を設定した局所解に現在状態を戻して探索を継続する局所解再探索手段と、を有することを特徴とする組合せ最適化問題の解析装置にある。   The present invention starts from an initial combination state, determines a state to be transitioned from among combination states defined as neighboring states using an evaluation function, and sequentially performs a search for repeating the transition to the transition destination. By means of a computer-aided combination optimization problem analysis apparatus for obtaining an optimal combination state for minimizing or maximizing an evaluation function, the evaluation function is expressed as a function with a plurality of partial evaluation function values, It is guaranteed that the evaluation function value does not increase or change when all the partial evaluation function values increase in the partial evaluation value area, and the best solution and the current combination in which the evaluation function value is the best combination state searched The best solution updating means that compares the evaluation function value of the current state that is the state and makes the current state the best solution when the evaluation function value of the current state is better, and the current state evaluates it Partial evaluation function improvement that sets a partial evaluation function improvement condition that any one of the partial evaluation function values is better than the local solution when the numerical value is a local solution better than the evaluation function values of all adjacent states When the condition setting means and the partial evaluation function improvement condition are set, the search range limiting means for searching only the adjacent state satisfying the partial evaluation function improvement condition among the adjacent states of the current state, and the local The partial evaluation function improvement condition cancellation means for canceling the partial evaluation function improvement condition when the evaluation function is in a better state than the solution, and when the partial evaluation function improvement condition is set, the current state becomes the local solution And a local solution re-search means for returning the current state to the local solution for which the partial evaluation function improvement condition is set and continuing the search.

この発明によれば、局所解脱出にあたって改善するべき部分評価関数を予め定めて、脱出用の探索を行うため、確率的な局所解脱出等、無目的な脱出方法よりも効率的な局所解脱出が可能となる。   According to the present invention, a partial evaluation function to be improved upon local solution escape is determined in advance, and a search for escape is performed. Therefore, a local solution escape more efficient than an unintended escape method such as stochastic local solution escape is performed. Is possible.

この発明によれば、局所解に陥った場合には、評価関数の構成要素である部分評価関数が、局所解よりも改善する状態だけに絞って探索を行うことにより局所解から脱出する、すなわち、局所解よりも評価関数値の小さな状態まで探索を進める。局所解脱出までの基本的な流れは、(1)まず、局所解よりも改善するべき部分評価関数Aを定め、(2)局所解脱出のための探索開始時点では、部分評価関数Aの改善に伴い、部分評価関数A以外の部分評価関数Bは一時悪化するが、(3)局所解脱出のための探索を継続することにより、悪化した部分評価関数Bが改善され、(4)現在状態の評価関数値が局所解の評価関数値よりも小さくなった時点で、局所解脱出を完了とする。以下この発明を実施の形態に従って説明する。   According to the present invention, when falling into a local solution, the partial evaluation function, which is a component of the evaluation function, escapes from the local solution by performing a search only for a state that improves over the local solution, that is, The search is advanced to a state where the evaluation function value is smaller than the local solution. The basic flow until the local solution escapes is as follows: (1) First, the partial evaluation function A to be improved over the local solution is determined, and (2) the partial evaluation function A is improved at the start of the search for the local solution escape. As a result, the partial evaluation function B other than the partial evaluation function A temporarily deteriorates, but (3) by continuing the search for local solution escape, the deteriorated partial evaluation function B is improved, and (4) the current state When the evaluation function value of becomes smaller than the evaluation function value of the local solution, the local solution escape is completed. Hereinafter, the present invention will be described according to embodiments.

実施の形態1.
図1はこの発明の一実施の形態による組合せ最適化問題の解析装置の構成を示す図である。この装置はコンピュータで構成され、本体部1にはCPU11、CPU11で使用するプロクラム、データをそれぞれ格納したプログラム部12a、データ部12bを含むメモリ12が設けられている。そして例えばデータ及び演算処理指示の入力を行う入力装置2等からの設定指示入力に従い、CPU11がメモリ12のプログラム部12aのアルゴリズムを記述したプログラムに従いデータ部12bの被解析データを処理して、例えば、1つ1つの組合せの状態の集まりである電力供給状態における様々な状態の組合せの中から最も良い組合せを求める最適化処理のための組合せ最適化問題の解析を行う。解析演算処理途中のデータや解析結果はデータ部12bに格納されまたこれらは表示装置3でモニタすることができる。なお、プログラム及びデータはメモリ12の代わりにディスク等(図示省略)に格納されていてもよい。
Embodiment 1 FIG.
FIG. 1 is a diagram showing the configuration of a combination optimization problem analyzing apparatus according to an embodiment of the present invention. This apparatus is constituted by a computer, and a main body 1 is provided with a memory 11 including a CPU 11, a program used by the CPU 11, a program part 12a storing data, and a data part 12b, respectively. Then, for example, in accordance with a setting instruction input from the input device 2 or the like that inputs data and arithmetic processing instructions, the CPU 11 processes the analyzed data in the data section 12b according to a program describing the algorithm of the program section 12a in the memory 12, A combination optimization problem is analyzed for an optimization process for obtaining the best combination from various combinations of states in the power supply state, which is a set of states of each combination. Data and analysis results in the middle of the analysis calculation process are stored in the data portion 12 b and can be monitored by the display device 3. The program and data may be stored on a disk or the like (not shown) instead of the memory 12.

そして図2〜図3は、この発明による組合せ最適化問題の解析装置の動作を示すフローチャートであり、評価関数を最小化する場合のものである。以下図に従って動作を説明すると、まず入力装置2から現在状態が初期設定されると、これに対する評価関数値および部分評価関数値を算出する(ステップS1)。ここで、評価関数は、複数の部分評価関数の総和であることを想定している。次に現在状態を最良解としてデータ部12bに保存する(ステップS2)。なお最良解とは、それまでの探索で求まった評価関数が最適な解をいう。   FIG. 2 to FIG. 3 are flowcharts showing the operation of the analysis apparatus for the combinatorial optimization problem according to the present invention, in the case where the evaluation function is minimized. The operation will be described below with reference to the drawings. First, when the current state is initialized from the input device 2, an evaluation function value and a partial evaluation function value are calculated (step S1). Here, it is assumed that the evaluation function is the sum of a plurality of partial evaluation functions. Next, the current state is stored in the data part 12b as the best solution (step S2). The best solution is a solution in which the evaluation function obtained by the search so far is optimum.

次に、極小解段数を0とする(ステップS3)。この例では、極小解から脱出するための探索を実施する過程で、さらに極小解に陥った場合に、最初の極小解を1段、次の極小解を2段とし、予め定めた最大の段数までは極小解のデータ部12bへの保存を行う処理構成としている。極小解段数が0は、まだ極小解に陥っていないことを意味する。   Next, the minimum number of steps is set to 0 (step S3). In this example, in the process of performing a search for escaping from a minimal solution, if the solution falls into a minimal solution, the first minimal solution is one step, the next minimal solution is two steps, and a predetermined maximum number of steps. Up to this point, the processing configuration is such that the minimal solution is stored in the data portion 12b. A minimum solution stage number of 0 means that the solution has not fallen into a minimum solution yet.

次に、現在状態の全ての隣の状態に順次着目し(ステップS4)、隣の状態の評価関数値および部分評価関数値を算出する(ステップS5)。   Next, attention is sequentially paid to all adjacent states in the current state (step S4), and the evaluation function value and partial evaluation function value of the adjacent state are calculated (step S5).

次に、極小解段数毎の部分評価関数改善条件を満たすかどうかを判定し(ステップS6)、満たしていればステップS7に進み、満たしていなければ着目している隣の状態を、現在状態の遷移対象からは外して、次の隣の状態に着目すべく、ステップS9へ進む。なお、極小解段数が0の場合は、すべての隣の状態が部分評価関数改善条件を満足するものとする。   Next, it is determined whether or not the partial evaluation function improvement condition for each minimum solution stage number is satisfied (step S6). If the condition is satisfied, the process proceeds to step S7. The process proceeds to step S9 to remove the transition target and focus on the next adjacent state. When the minimum number of solution stages is 0, all adjacent states satisfy the partial evaluation function improvement condition.

極小解段数毎の部分評価関数改善条件について説明すると、本発明では、極小解から脱出するにあたり、脱出のために定めた部分評価関数値もしくは複数の部分評価関数値の合計値が極小解よりも改善される状態のみに絞りこんで探索を行うが、この絞込み条件を、その極小解の部分評価関数改善条件とする。例えば、部分評価関数A1に対する部分評価関数改善条件では、部分評価関数A1の値が極小解での値よりも小さいことが必要になる。部分評価関数A1と部分評価関数A2の合計に対する部分評価関数改善条件では、部分評価関数A1と部分評価関数A2の和が極小解での値よりも小さいことが必要であり、部分評価関数A1または部分評価関数A2のどちらかが極小解での値よりも大きくなってもよい。なお、複数の部分評価関数値の合計値に対する部分評価関数改善条件は、評価関数が部分評価関数値の和である場合だけ使用する。本例では複数段の極小解にわたって状態の絞込みを行うため、現在の極小解段数以下の全ての極小解についての極小解脱出のための部分評価関数改善条件を満たす必要があり、これを、極小解段数毎の部分評価関数改善条件と記述している。   The conditions for improving the partial evaluation function for each number of minimum solution stages will be described.In the present invention, when escaping from the minimum solution, the partial evaluation function value or the total value of the plurality of partial evaluation function values determined for escaping is less than the minimum solution. The search is performed by narrowing down only to the improved state, and this narrowing condition is set as the partial evaluation function improvement condition of the minimal solution. For example, in the partial evaluation function improvement condition for the partial evaluation function A1, the value of the partial evaluation function A1 needs to be smaller than the value in the minimum solution. In the partial evaluation function improvement condition with respect to the sum of the partial evaluation function A1 and the partial evaluation function A2, the sum of the partial evaluation function A1 and the partial evaluation function A2 needs to be smaller than the value in the minimum solution, and the partial evaluation function A1 or Either of the partial evaluation functions A2 may be larger than the value at the minimum solution. The partial evaluation function improvement condition for the total value of the plurality of partial evaluation function values is used only when the evaluation function is the sum of the partial evaluation function values. In this example, since the state is narrowed down over a plurality of minimum solutions, it is necessary to satisfy the partial evaluation function improvement conditions for exiting the minimum solution for all the minimum solutions below the current minimum solution stage number. It is described as a partial evaluation function improvement condition for each solution stage number.

次に、極小解脱出用隣接状態の候補を選出する(ステップS7)。この処理は、現在状態が極小解と判定された場合の極小解脱出用隣接状態の保存に備えて、予めその候補を選出しておく処理である。現在状態よりも隣の状態の方が値の良い部分評価関数についてのみ、部分評価関数値の改善量を合計し、これが所定値よりも大きければ極小解脱出用隣接状態の候補とする。   Next, an adjacent state candidate for minimum solution escape is selected (step S7). This process is a process in which candidates are selected in advance in preparation for storing the adjacent state for minimum solution escape when it is determined that the current state is the minimum solution. Only for the partial evaluation function whose value is better in the state adjacent to the current state, the improvement amount of the partial evaluation function value is summed up, and if this is larger than a predetermined value, it is determined as a candidate for the adjacent state for minimum solution escape.

次に、評価関数値の最も良い隣の状態をデータ部12bに保存する(ステップS8)。次にステップS9では、全ての隣の状態に着目したかどうかを判定し、着目済みであれば図3のステップS10へ進み、そうでなければステップS4へ戻って次の隣の状態に着目する。   Next, the next state with the best evaluation function value is stored in the data part 12b (step S8). Next, in step S9, it is determined whether or not attention has been paid to all adjacent states. If attention has already been paid, the process proceeds to step S10 in FIG. 3, and if not, the process returns to step S4 to focus on the next adjacent state. .

次に、ステップS10では、評価関数値の最も良い隣の状態よりも現在状態の方が評価関数値が良いか同じであれば、現在状態を極小解と判定し、ステップS11へ進む。現在状態が極小解でなければ、ステップS19へ進む。ステップS11では、極小解段数が既に最大値に達しているかどうかを判定し、最大に達していなければステップS12へ進み、最大に達していればステップS13へ進む。   Next, in step S10, if the evaluation function value is better or the same in the current state than the next state having the best evaluation function value, the current state is determined as the minimum solution, and the process proceeds to step S11. If the current state is not the minimum solution, the process proceeds to step S19. In step S11, it is determined whether or not the minimum number of steps has already reached the maximum value. If it has not reached the maximum value, the process proceeds to step S12, and if it has reached the maximum, the process proceeds to step S13.

ステップS12では、極小解段数をカウントアップし、カウントアップ後の極小解段数での、極小解、極小解の評価関数値、極小解の各部分評価関数値、極小解脱出用隣接状態をデータ部12bに保存する。極小解脱出用隣接状態は、ステップS7で選出した極小解脱出用隣接状態の候補から、評価関数値の良い順に例えば上位5つの隣の状態を選出し、カウントアップ後の極小解段数からの脱出用隣接状態としてデータ部12bに保存する。ステップS12の実行後は、ステップS17に進む。   In step S12, the number of minimum solution steps is counted up, and the minimum solution, the evaluation function value of the minimum solution, each partial evaluation function value of the minimum solution, and the adjacent state for the minimum solution escape are stored in the data portion. Save to 12b. As the adjacent state for minimum solution escape, for example, the top five neighboring states are selected from the candidates for the minimum solution escape adjacent state selected in step S7 in the order of good evaluation function values, and the escape from the minimum solution stage number after counting up is selected. And stored in the data portion 12b as the adjacent state. After execution of step S12, the process proceeds to step S17.

ステップS13では、現在の段数の極小解に現在状態を戻し、着目している極小解脱出用隣接状態を削除する。ステップS14では、次に着目する極小解脱出用隣接状態があるかどうかをチェックする。ステップS14で極小解脱出用隣接状態が無い場合、ステップS15で現在の極小解段数が1か否かを判断し、1であれば、すなわち極小解段数1段の極小解脱出用隣接状態が1つもなくなった場合に探索を終了し、そうでない場合にはステップS16に進んで極小解段数をカウントダウンしステップS13に戻る。ステップS14で極小解脱出用隣接状態が有る場合は、ステップS17に進む。   In step S13, the current state is returned to the minimum solution of the current number of steps, and the focused minimum solution escape adjacent state is deleted. In step S14, it is checked whether there is an adjacent state for minimal solution escape to be focused next. If there is no minimum solution escape adjacent state in step S14, it is determined in step S15 whether or not the current minimum solution step number is 1, if it is 1, that is, the minimum solution escape adjacent state of 1 step is 1. The search is terminated when it is lost, and if not, the process proceeds to step S16 to count down the minimum number of steps and return to step S13. If there is an adjacent state for minimum solution escape in step S14, the process proceeds to step S17.

次にステップS17では、現在の極小解段数から脱出するために、次に着目する極小解脱出用隣接状態を決定する。決定のための優先順位はステップS12において既に決めてあるため、これに従う。   Next, in step S17, in order to escape from the current minimum solution stage number, the adjacent state for minimum solution escape to be focused next is determined. Since the priority order for determination is already determined in step S12, this is followed.

次にステップS18では、現在状態を着目した極小解脱出用隣接状態に遷移させ、この遷移に伴い改善した部分評価関数を、現在の極小解段数に対する改善対象部分評価関数とする。この改善対象部分評価関数は、現在の段数の極小解から脱出、すなわち極小解段数のカウントダウンを行うまで、探索の対象となる状態を選択する条件となる。   Next, in step S18, the current state is shifted to an adjacent state for minimum solution escape, and the partial evaluation function improved with this transition is set as a partial evaluation function to be improved with respect to the current minimum number of solution stages. This improvement target partial evaluation function is a condition for selecting a state to be searched until the current minimum number of steps is escaped, that is, until the number of minimum solution steps is counted down.

次にステップS19では、現在状態を評価関数値が最も良い隣の状態に遷移させる。次にステップS20では、現在状態の評価関数値および部分評価関数値を算出する。次にステップS21では、現在状態の評価関数値が最良解より良ければ現在状態を最良解としてデータ部12bに更新保存する。次にステップS22では、探索を所定回数実施済みであれば探索を終了し、そうでなければステップS23へ進む。   Next, in step S19, the current state is shifted to the next state having the best evaluation function value. Next, in step S20, an evaluation function value and a partial evaluation function value in the current state are calculated. In step S21, if the evaluation function value in the current state is better than the best solution, the current state is updated and stored in the data unit 12b as the best solution. Next, in step S22, if the search has been performed a predetermined number of times, the search is terminated; otherwise, the process proceeds to step S23.

そしてステップS23では、極小解段数解除処理を行い、図2のステップS4へ戻る。極小解段数解除処理は、現在状態の評価関数値が、現在の極小解段数の極小解の評価関数値よりも良ければ、現在の段数の極小解を脱出したと判定し、極小解段数をカウントダウンする。一度で複数段の極小解脱出が可能な場合もあるため、極小解段数解除処理は、現在の極小解段数から0段に向かって、可能な限りカウントダウンを行う。   In step S23, the minimum solution stage number cancellation process is performed, and the process returns to step S4 in FIG. The minimum number of steps cancellation process determines that the minimum solution of the current number of steps has been escaped if the evaluation function value of the current state is better than the evaluation function value of the minimum number of solutions of the current minimum number of steps, and the number of minimum solutions is counted down. To do. Since there may be a case where a plurality of minimum solution escapes are possible at one time, the minimum solution step number cancellation processing counts down as much as possible from the current minimum solution step number toward zero.

なお、これらの解析演算処理途中のデータや解析結果は表示装置3に表示することでモニタすることができる。   Note that data and analysis results in the middle of the analysis calculation processing can be monitored by displaying them on the display device 3.

また上記実施の形態において、評価関数および部分評価関数は評価関数値が小さいものを評価がよいとしているが、評価関数値が大きいものを評価がよいとする評価関数および部分評価関数を使用してもよく、さらに評価関数および部分評価関数の符号を反転させれば、最大化の組合せ問題が最小化の組合せ問題に、また最小化の組合せ問題が最大化の組合せ問題に逆転し、いずれの場合においても本願発明は実施可能である。
また評価関数は部分評価関数の和だけでなく、すべての部分評価値の領域において、すべての部分評価関数値が増加すると評価関数値が増加または変化しないことが保証されている部分評価関数値の合成関数であればよい。この条件は、評価関数の最小化でも最大化でも同じである。
In the above embodiment, the evaluation function and the partial evaluation function are good when the evaluation function value is small, but the evaluation function and the partial evaluation function are used when the evaluation function value is high. In addition, if the sign of the evaluation function and the partial evaluation function is reversed, the combination problem of maximization is reversed to the combination problem of minimization, and the combination problem of minimization is reversed to the combination problem of maximization. However, the present invention can be implemented.
In addition, the evaluation function is not only the sum of partial evaluation functions, but in the area of all partial evaluation values, the partial evaluation function values for which it is guaranteed that the evaluation function values do not increase or change when all the partial evaluation function values increase. Any composite function may be used. This condition is the same whether the evaluation function is minimized or maximized.

この発明の一実施の形態による組合せ最適化問題の解析装置の構成を示す図である。It is a figure which shows the structure of the analysis apparatus of the combination optimization problem by one Embodiment of this invention. この発明による組合せ最適化問題の解析装置の動作を示すフローチャートである。It is a flowchart which shows operation | movement of the analysis apparatus of the combination optimization problem by this invention. 図2に続く動作を示すフローチャートである。3 is a flowchart illustrating an operation following FIG.

符号の説明Explanation of symbols

1 本体部、2 入力装置、3 表示装置、11 CPU、12 メモリ、12a プログラム部、12b データ部。   DESCRIPTION OF SYMBOLS 1 Main part, 2 Input device, 3 Display apparatus, 11 CPU, 12 Memory, 12a Program part, 12b Data part.

Claims (4)

初期の組合せ状態から出発し、隣の状態と定義された組合せ状態の中から評価関数を用いて遷移すべき状態を決定し、順次この遷移先への遷移を繰り返す探索を行うことにより、評価関数を最小化または最大化する最適な組合せ状態を求めるコンピュータを用いた組合せ最適化問題の解析装置であって、
前記評価関数が複数の部分評価関数値による関数として表現され、すべての部分評価値の領域においてすべての部分評価関数値が増加すると評価関数値が増加または変化しないことが保証されており、
探索した中で評価関数値が最もよい組合せ状態である最良解と現在の組合せ状態である現在状態の評価関数値を比較し、現在状態の評価関数値の方がよい場合に現在状態を最良解とする最良解更新手段と、
現在状態がその評価関数値がすべての隣接状態の評価関数値よりもよい局所解である場合に、局所解よりも何れか1個の部分評価関数値がよいことである部分評価関数改善条件を設定する部分評価関数改善条件設定手段と、
部分評価関数改善条件が設定されている場合は、現在状態の隣接状態の中で部分評価関数改善条件を満足する隣接状態だけを探索の対象とする探索範囲限定手段と、
局所解よりも評価関数がよい状態に遷移できた際に部分評価関数改善条件を解除する部分評価関数改善条件解除手段と、
部分評価関数改善条件が設定されている場合に現在状態が局所解になった場合に、部分評価関数改善条件を設定した局所解に現在状態を戻して探索を継続する局所解再探索手段と、
を有することを特徴とする組合せ最適化問題の解析装置。
An evaluation function starts from an initial combination state, determines a state to be transitioned from among the combination states defined as neighboring states by using an evaluation function, and sequentially repeats the transition to the transition destination. A computer-aided analysis apparatus for combinatorial optimization problems that seeks an optimal combination state for minimizing or maximizing
The evaluation function is expressed as a function with a plurality of partial evaluation function values, and it is guaranteed that the evaluation function value does not increase or change when all the partial evaluation function values increase in the area of all the partial evaluation values,
Compare the best solution that is the best combination state with the best evaluation function value in the search and the evaluation function value of the current state that is the current combination state, and if the evaluation function value of the current state is better, And the best solution update means
If the current state is a local solution whose evaluation function value is better than the evaluation function values of all adjacent states, the partial evaluation function improvement condition is that any one partial evaluation function value is better than the local solution. A partial evaluation function improvement condition setting means to be set;
When the partial evaluation function improvement condition is set, the search range limiting means for searching only the adjacent state satisfying the partial evaluation function improvement condition among the adjacent states of the current state,
A partial evaluation function improvement condition canceling means for canceling the partial evaluation function improvement condition when the evaluation function is in a better state than the local solution;
When the partial evaluation function improvement condition is set, when the current state becomes a local solution, the local solution re-search means for returning the current state to the local solution in which the partial evaluation function improvement condition is set and continuing the search,
A device for analyzing a combinatorial optimization problem, characterized by comprising:
部分評価関数改善条件が設定されている場合に現在状態が局所解になった時は、局所解の数が1より大きい所定個以下であれば、新たな局所解に対する部分評価関数改善条件を設定して探索を継続し、部分評価関数改善条件を局所解ごとに管理することを特徴とする請求項1に記載の組合せ最適化問題の解折装置。   When partial evaluation function improvement conditions are set, if the current state becomes a local solution, if the number of local solutions is less than or equal to a predetermined number, set partial evaluation function improvement conditions for a new local solution The combinatorial optimization problem solving apparatus according to claim 1, wherein the search is continued and the partial evaluation function improvement condition is managed for each local solution. 評価関数を部分評価関数値の和とすることを特徴とする請求項2に記載の組合せ最適化問題の解析装置。   3. The combination optimization problem analyzing apparatus according to claim 2, wherein the evaluation function is a sum of partial evaluation function values. 部分評価関数改善条件として複数の部分評価関数値の和が局所解よりもよいという条件も使用することを特徴とする請求項3に記載の組合せ最適化問題の解析装置。   4. The combination optimization problem analyzing apparatus according to claim 3, wherein a condition that a sum of a plurality of partial evaluation function values is better than a local solution is used as the partial evaluation function improvement condition.
JP2004257120A 2004-09-03 2004-09-03 Device for analyzing combination optimization problem Pending JP2006072820A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2004257120A JP2006072820A (en) 2004-09-03 2004-09-03 Device for analyzing combination optimization problem

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2004257120A JP2006072820A (en) 2004-09-03 2004-09-03 Device for analyzing combination optimization problem

Publications (1)

Publication Number Publication Date
JP2006072820A true JP2006072820A (en) 2006-03-16

Family

ID=36153363

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2004257120A Pending JP2006072820A (en) 2004-09-03 2004-09-03 Device for analyzing combination optimization problem

Country Status (1)

Country Link
JP (1) JP2006072820A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8314571B2 (en) 2010-12-14 2012-11-20 Greenwave Reality, Pte, Ltd. Light with changeable color temperature
US8764210B2 (en) 2010-07-19 2014-07-01 Greenwave Reality Pte Ltd. Emitting light using multiple phosphors
US8820981B2 (en) 2010-07-19 2014-09-02 Greenwave Reality Pte Ltd Electrically controlled glass in a lamp
JP2015148926A (en) * 2014-02-06 2015-08-20 富士通株式会社 Information processing apparatus, evaluation function learning method, and program
CN106021850A (en) * 2016-05-05 2016-10-12 江苏建筑职业技术学院 Entropy weight TOPSIS method based earth excavation machine combination analysis method
CN110595008A (en) * 2019-10-28 2019-12-20 山东建筑大学 Multi-equipment collaborative optimization method and system for ground source heat pump air conditioning system
EP3975032A1 (en) 2020-04-07 2022-03-30 Fujitsu Limited Optimization apparatus, optimization method, and optimization program

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8764210B2 (en) 2010-07-19 2014-07-01 Greenwave Reality Pte Ltd. Emitting light using multiple phosphors
US8820981B2 (en) 2010-07-19 2014-09-02 Greenwave Reality Pte Ltd Electrically controlled glass in a lamp
US8314571B2 (en) 2010-12-14 2012-11-20 Greenwave Reality, Pte, Ltd. Light with changeable color temperature
JP2015148926A (en) * 2014-02-06 2015-08-20 富士通株式会社 Information processing apparatus, evaluation function learning method, and program
CN106021850A (en) * 2016-05-05 2016-10-12 江苏建筑职业技术学院 Entropy weight TOPSIS method based earth excavation machine combination analysis method
CN110595008A (en) * 2019-10-28 2019-12-20 山东建筑大学 Multi-equipment collaborative optimization method and system for ground source heat pump air conditioning system
EP3975032A1 (en) 2020-04-07 2022-03-30 Fujitsu Limited Optimization apparatus, optimization method, and optimization program

Similar Documents

Publication Publication Date Title
CN108491226B (en) Spark configuration parameter automatic tuning method based on cluster scaling
Tseng et al. Multiple trajectory search for unconstrained/constrained multi-objective optimization
US11003826B1 (en) Automated analysis and optimization of circuit designs
Li et al. A hybrid Pareto-based local search algorithm for multi-objective flexible job shop scheduling problems
Zhang et al. Bottleneck machine identification method based on constraint transformation for job shop scheduling with genetic algorithm
JP6931624B2 (en) Learning support device and learning support method
Alhindi et al. MOEA/D with tabu search for multiobjective permutation flow shop scheduling problems
CN109543247B (en) NSGA-II-based analog integrated circuit parameter optimization design method and device
KR20150084596A (en) The method for parameter investigation to optimal design
JP2003223322A (en) Device for analyzing combinatorial optimization problem
JP2019046031A (en) Optimal solution search method, optimal solution search program, and optimal solution search apparatus
JP2006072820A (en) Device for analyzing combination optimization problem
JP4165712B2 (en) Data flow graph same subgraph detection device, high-level synthesis device, data flow graph same subgraph detection method, data flow graph same subgraph detection control program, and readable recording medium
JP2006252011A (en) Environmental load evaluation device, environmental load evaluation method and program recording medium
JP5966836B2 (en) Evaluation support method, information processing apparatus, and program
JPWO2014087590A1 (en) Optimization device, optimization method, and optimization program
Bibai et al. An evolutionary metaheuristic based on state decomposition for domain-independent satisficing planning
JP7508842B2 (en) Optimization device, optimization method, and optimization program
JP5964781B2 (en) SEARCH DEVICE, SEARCH METHOD, AND SEARCH PROGRAM
Lin et al. A new density-based scheme for clustering based on genetic algorithm
Glover Multi-wave algorithms for metaheuristic optimization
JP6613937B2 (en) Quality prediction apparatus, quality prediction method, program, and computer-readable recording medium
Lin et al. A TABU SEARCH ALGORITHM TO MINIMIZE TOTAL WEIGHTED TARDINESS FOR THE JOB SHOP SCHEDULING PROBLEM.
JP2007072883A (en) Cache memory analysis method, processor and simulative information processing device
JP2009265867A (en) Combined optimum solution calculation device

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20061016

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20090526

A02 Decision of refusal

Free format text: JAPANESE INTERMEDIATE CODE: A02

Effective date: 20091006