Solution search device and solution search method
The solution search device addresses the inefficiency of conventional multi-objective optimization methods by allowing direct refinement of solutions within the Pareto frontier, thereby reducing the time to find a satisfactory solution for decision makers.
Patent Information
- Application Number
- PCT/JP2023/019037
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-23
- Publication Date
- 2024-11-28
AI Technical Summary
Conventional methods for multi-objective optimization problems are time-consuming as they require repeated iterations to find an optimal solution that satisfies the decision maker, involving multiple steps of setting and adjusting aspiration levels and desired levels.
A solution search device that includes a setting unit for initial aspiration levels, an optimal solution calculation unit, and an adjustment unit to refine solutions within the Pareto frontier based on user input, allowing for direct adjustment of solutions to meet criteria more efficiently.
This approach significantly reduces the time required to obtain an optimal solution that satisfies the decision maker by enabling direct refinement of initial solutions within the Pareto frontier, streamlining the process compared to conventional methods.
Smart Images

Figure JP2023019037_28112024_PF_FP_ABST
Abstract
Description
Solution search device and solution search method
[0001] The present disclosure relates to a solution search device and a solution search method.
[0002] In multi-objective optimization problems where multiple objective functions are optimized, it is difficult to express all of the objective functions by aggregating them into an evaluation function such as a linear weighted sum. Therefore, various methods have been proposed for obtaining optimal solutions to multi-objective optimization problems. For example, Patent Literature 1 discloses a predictive control method using a satisficing trade-off method for multi-objective optimization problems. The satisficing trade-off method is an interactive multi-objective programming method that uses aspiration levels, and is one method for formulating multi-objective optimization problems.
[0003] The predictive control method described in Patent Document 1 (hereinafter also referred to as the "conventional method") accepts input of aspiration levels, which are target values for multiple objective functions (controlled quantities), to set the aspiration levels, calculates an optimal solution close to the aspiration levels, displays the optimal solution, accepts input to change the aspiration levels if the optimal solution does not satisfy a specified criterion, changes the setting of the aspiration levels, and repeats these steps to determine an optimal solution that satisfies the specified criterion.
[0004] Specifically, in the conventional method, the first optimal solution is calculated as the intersection of a line connecting the ideal point, which is the minimum value of the objective function, and a first aspiration level set by a decision maker (operator), with a line representing the Pareto frontier, which is a trade-off set of optimal solutions. If the decision maker is not satisfied with this first optimal solution, the aspiration level is changed from the first aspiration level to a second aspiration level through a trade-off analysis using sensitivity analysis, and the second optimal solution is calculated as the intersection of a line connecting this second aspiration level and the ideal point with the line representing the Pareto frontier.
[0005] JP 2010-152767 A
[0006] However, as described above, in conventional methods, if the decision maker is not satisfied with the calculated optimal solution, the aspiration level setting is changed and a different optimal solution is calculated based on the changed aspiration level, and this process is repeated until the decision maker is satisfied with the calculated optimal solution, which has the problem that it takes a long time to obtain an optimal solution that satisfies the decision maker.
[0007] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a solution search device that can shorten the time required to obtain an optimal solution that satisfies a decision maker compared to conventional methods.
[0008] A solution search device according to the present disclosure includes: a setting unit that sets a first aspiration level, which is a target value for multiple objective functions in a multi-objective optimization problem; a first optimal solution calculation unit that calculates a first optimal solution using the first aspiration level set by the setting unit and a frontier, which is a solution set for optimizing the multiple objective functions; and a first optimal solution adjustment unit that, when receiving an instruction from a user not to adopt the first optimal solution calculated by the first optimal solution calculation unit, receives an input of an adjustment amount for adjusting the first optimal solution from the user, and adjusts the first optimal solution within the range of the frontier based on the received adjustment amount.
[0009] According to the present disclosure, it is possible to reduce the time required to obtain an optimal solution that satisfies a decision maker compared to conventional methods.
[0010] 9A and 9B are diagrams showing an example of the hardware configuration of a solution search device according to embodiment 1. FIG. 9B is a diagram showing an example of the configuration of a solution search device according to embodiment 1. FIG. 9C is a flowchart showing an example of the operation of the solution search device according to embodiment 1. FIG. 9D is a diagram showing an example of extraction of a Pareto frontier by a solution extraction unit according to embodiment 1. FIG. 9E is a diagram showing an example of calculation of a first optimal solution according to embodiment 1. FIG. 9F is a diagram showing an example of calculation of a second optimal solution and a second aspiration level according to embodiment 1. FIG. 9G is a diagram showing details of an example of calculation of a second optimal solution and a second aspiration level according to embodiment 1. FIG. 9H is a diagram showing details of an example of calculation of a first optimal solution and a second optimal solution in a case where the solution space is a three-dimensional space according to embodiment 1. FIG. 9H is a diagram showing details of an example of calculation of a second aspiration level in a case where the solution space is a three-dimensional space according to embodiment 1. Fig. 10 is a diagram showing an example of calculation of a second optimal solution and a second aspiration level in embodiment 4. Fig. 11 is a diagram for explaining a reference region and an extension region in embodiment 5. Fig. 12 is a diagram for explaining a search region in embodiment 5.
[0011] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings.
[0012] Fig. 1 is a diagram showing an example of the configuration of a solution search device 1 according to embodiment 1. As shown in Fig. 1, the solution search device 1 includes, for example, a population generation unit 10, a solution extraction unit 11, an aspiration level / ideal point setting unit 12, a first optimal solution calculation unit 13, a first optimization calculation unit 14, a first optimal solution adjustment unit 15, a second aspiration level calculation unit 16, a second optimization calculation unit 17, a determination unit 18, an interface unit 20 (an input unit 21 and a display unit 22), a display control unit 23, and a screen generation unit 24.
[0013] The population generation unit 10 generates a set of solutions (population) for a multi-objective optimization problem in which multiple objective functions are optimized. For example, the population generation unit 10 uses a known optimization algorithm such as NSGA-II (Non-Dominated Sorting Genetic Algorithm II) to generate a set of solutions (population) in a space defined by multiple objective functions (hereinafter also referred to as a "solution space").
[0014] The solution extraction unit 11 extracts a frontier, which is a set of solutions for optimizing multiple objective functions, from the set of solutions generated by the population generation unit 10. For example, the solution extraction unit 11 extracts a Pareto frontier from the set of solutions generated by the population generation unit 10 using a known optimization algorithm such as the above-mentioned NSGA-II.
[0015] The Pareto frontier is a set of Pareto solutions. A Pareto solution is a solution in which optimizing one of multiple objective functions results in a deterioration of the other objective functions. When there are two objective functions, i.e., when the solution space is two-dimensional, the Pareto frontier is represented by a line connecting the Pareto solutions in the solution space. When there are three objective functions, i.e., when the solution space is three-dimensional, the Pareto frontier is represented by a surface including the Pareto solutions in the solution space.
[0016] A decision maker, who is a user of the solution search device 1, inputs an ideal point and a first aspiration level to the solution search device 1 via the interface unit 20. The aspiration level / ideal point setting unit 12 accepts the ideal point and first aspiration level input by the decision maker via the interface unit 20, and sets the accepted ideal point and first aspiration level in the solution space.
[0017] The ideal point is a point that represents an optimal solution (e.g., a minimum value) for all of the objective functions. The aspiration level is a point that represents a minimum standard that should be satisfied for each of the objective functions, and the first aspiration level is the aspiration level that the decision maker first inputs (specifies).
[0018] The first optimal solution calculation unit 13 calculates a first optimal solution using the ideal point and first aspiration level set by the aspiration level / ideal point setting unit 12 and the frontier (e.g., Pareto frontier) extracted by the solution extraction unit 11.
[0019] For example, the first optimal solution calculation unit 13 identifies an intersection in the solution space between a line connecting the ideal point set by the aspiration level / ideal point setting unit 12 and the first aspiration level and a line indicating the frontier extracted by the solution extraction unit 11, and calculates the identified intersection as the first optimal solution. The first optimal solution calculated here is presented to the decision maker via the interface unit 20.
[0020] The decision maker checks the presented first optimal solution and makes the following judgments. The decision maker then inputs the judgment results to the solution search device 1 via the interface unit 20. (1) Whether to adopt the presented first optimal solution. (2) If the presented first optimal solution is not adopted, whether to optimize the first optimal solution based on the first aspiration level. (3) If the presented first optimal solution is not adopted and the first optimal solution is not optimized based on the first aspiration level, whether to adjust the first optimal solution within the range of the frontier. Note that if the judgment result in (1) above indicates that the presented first optimal solution is to be adopted, the solution search device 1 ends the processing at that stage.
[0021] If the determination result in (2) indicates that the presented first optimal solution should not be adopted and that the first optimal solution should be optimized based on the first aspiration level, the first optimization calculation unit 14 searches for a solution different from the first optimal solution that is located in the vicinity of the first optimal solution in the solution space based on the determination result. The first optimization calculation unit 14 presents the searched solution to the decision maker via the interface unit 20, and the decision maker determines whether to adopt the presented solution. Thereafter, the first optimization calculation unit 14 repeats the search for the solution until the presented solution is adopted by the decision maker. Details of the solution search by the first optimization calculation unit 14 will be described in embodiments 5 and 6.
[0022] If the determination result in (3) indicates that the presented first optimal solution should not be adopted, that the first optimal solution should not be optimized based on the first aspiration level, and that the first optimal solution should be adjusted within the range of the frontier, the first optimal solution adjustment unit 15 adjusts the first optimal solution within the range of the frontier based on the determination result, thereby calculating a second optimal solution different from the first optimal solution.
[0023] For example, when the solution space is a two-dimensional space, the first optimal solution adjustment unit 15 moves the first optimal solution calculated by the first optimal solution calculation unit 13 on the line indicating the frontier. At this time, the first optimal solution adjustment unit 15 receives an adjustment amount (movement amount) of the first optimal solution from the decision maker, and adjusts (moves) the first optimal solution on the line indicating the frontier based on the received adjustment amount.
[0024] Furthermore, for example, when the solution space is a three-dimensional space, the first optimal solution adjustment unit 15 moves the first optimal solution calculated by the first optimal solution calculation unit 13 on a plane representing the frontier. At this time, the first optimal solution adjustment unit 15 receives an adjustment amount (movement amount) of the first optimal solution from the decision maker, and adjusts (moves) the first optimal solution on the plane representing the frontier based on the received adjustment amount.
[0025] The first optimal solution (i.e., the second optimal solution) adjusted as described above is presented to the decision maker via the interface unit 20.
[0026] The second aspiration level calculation unit 16 calculates a second aspiration level different from the first aspiration level based on the first optimal solution (i.e., the second optimal solution) adjusted by the first optimal solution adjustment unit 15 and the ideal point that indicates the optimal solutions for all of the objective functions among the plurality of objective functions. The second aspiration level calculated here is presented to the decision maker via the interface unit 20.
[0027] The decision maker checks the presented second optimal solution and second aspiration level and makes the following judgments. The decision maker then inputs the judgment results to the solution search device 1 via the interface unit 20. (4) Whether to adopt the presented second optimal solution; (5) If the presented second optimal solution is not adopted, whether to optimize the second optimal solution based on the second aspiration level; (6) If the presented second optimal solution is not adopted and the second optimal solution is not optimized based on the second aspiration level, whether to start over from setting the first aspiration level. Note that if the judgment result in (4) above indicates that the presented second optimal solution is to be adopted, the solution search device 1 ends the processing at that stage.
[0028] If the determination result in (5) indicates that the presented second optimal solution should not be adopted and that the second optimal solution should be optimized based on the second aspiration level, the second optimization calculation unit 17 searches for a solution that is different from the second optimal solution and is located in the vicinity of the second optimal solution in the solution space based on the determination result. The second optimization calculation unit 17 presents the searched solution to the decision maker via the interface unit 20, and the decision maker determines whether to adopt the presented solution. Thereafter, the second optimization calculation unit 17 repeats the search for the solution until the presented solution is adopted by the decision maker. Details of the solution search by the second optimization calculation unit 17 will be described in embodiments 5 and 6.
[0029] If the judgment result in (6) indicates that the presented second optimal solution will not be adopted, that the second optimal solution will not be optimized based on the second aspiration level, and that the first aspiration level will not be reset, the first optimal solution adjustment unit 15 adjusts (readjusts) the second optimal solution within the range of the frontier based on the judgment result. As a result, the first optimal solution adjustment unit 15 calculates a new second optimal solution that is different from the previously calculated second optimal solution. At this time, the first optimal solution adjustment unit 15 again receives an adjustment amount (movement amount) from the decision maker, and adjusts (moves) the second optimal solution within the range of the frontier based on the received adjustment amount.
[0030] Furthermore, if the judgment result in (6) indicates that the presented second optimal solution should not be adopted, that the second optimal solution should not be optimized based on the second aspiration level, and that the process should be started over from setting the first aspiration level, the decision maker inputs an aspiration level different from the previously input first aspiration level as a new first aspiration level to the solution search device 1 via the interface unit 20. The aspiration level / ideal point setting unit 12 accepts the new first aspiration level input by the decision maker via the interface unit 20 and sets the accepted new first aspiration level in the solution space. Thereafter, the first optimal solution calculation unit 13 recalculates the first optimal solution based on the new first aspiration level, as described above.
[0031] The determination unit 18 determines the content of each of the above-mentioned determination results input by the decision maker.
[0032] The interface unit 20 provides an interface function for the user, i.e., the decision maker, and includes an input unit 21 including, for example, a keyboard and a mouse, and a display unit 22 including a display.
[0033] The display control unit 23 causes the display unit 22 to display a screen indicated by the data generated by the screen generation unit 24 .
[0034] The screen generation unit 24 generates data showing a predetermined screen for the decision maker to input the ideal point, first aspiration level, adjustment amount, etc., and to confirm the first optimal solution and second optimal solution, etc.
[0035] Next, an example of the operation of the solution search device 1 according to the first embodiment will be described with reference to the flowchart shown in FIG.
[0036] For simplicity, the following description will be given by taking as an example a case where an optimal solution is calculated by the solution search device 1 for a multi-objective optimization problem with two objective functions. The following description will also be given by taking as an example a case where the solution extraction unit 11 extracts a Pareto frontier as a frontier using a known optimization algorithm such as the above-mentioned NSGA-II.
[0037] First, the population generation unit 10 uses the above-described optimization algorithm to generate a set (population) of solutions to a multi-objective optimization problem that optimizes two objective functions (step ST1).
[0038] Next, the solution extractor 11 uses the above-mentioned optimization algorithm to extract a Pareto frontier from the set of solutions generated by the population generator 10 (step ST2).
[0039] An example of the Pareto frontier extracted by the solution extractor 11 is shown in Fig. 3. In Fig. 3, the horizontal axis represents the first objective function (g1) and the vertical axis represents the second objective function (g2). In this case, the solution space K is a two-dimensional space.
[0040] 3, points A to J indicate a set of solutions (population) generated by the population generation unit 10, and among these, points A to F indicate Pareto solutions. In FIG. 3, line PF connecting points A to F indicates the Pareto frontier (set of Pareto solutions). Note that points G to J indicate those of the set of solutions (population) generated by the population generation unit 10 that are not Pareto solutions. Note that in the following explanation, for convenience of explanation, points A to F that are Pareto solutions will also be referred to as "solutions on the Pareto frontier."
[0041] Next, the decision maker inputs the ideal point and the first aspiration level to the solution search device 1 via the interface unit 20. The aspiration level / ideal point setting unit 12 accepts the ideal point and the first aspiration level input by the decision maker via the interface unit 20, and sets the accepted ideal point and first aspiration level in the solution space K (step ST3).
[0042] Next, the first optimal solution calculation unit 13 calculates a first optimal solution using the ideal point and first aspiration level set by the aspiration level / ideal point setting unit 12 and the Pareto frontier extracted by the solution extraction unit 11 (step ST4).
[0043] An example of the first optimum solution calculated by the first optimum solution calculation unit 13 is shown in Fig. 4. In Fig. 4, point P1 indicates the ideal point, point Q1 indicates the first aspiration level, and point R indicates the first optimum solution.
[0044] For example, the first optimal solution calculation unit 13 identifies the intersection of a line connecting the ideal point P1 set by the aspiration level / ideal point setting unit 12 and the first aspiration level Q1 with the line PF indicating the Pareto frontier extracted by the solution extraction unit 11, and calculates the identified intersection as the first optimal solution R. The first optimal solution R is a solution on the Pareto frontier. The first optimal solution R calculated here is presented to the decision maker via the interface unit 20.
[0045] Next, the decision maker checks the presented first optimal solution R, judges the above points (1) to (3), and inputs the judgment results to the solution search device 1 via the interface unit 20.
[0046] Next, the determination unit 18 determines whether the determination result indicates that the first optimal solution R presented in (1) above should be adopted (step ST5). As a result, if it is determined that the determination result indicates that the first optimal solution R presented in (1) above should be adopted (step ST5; YES), the processing ends. On the other hand, if it is determined that the determination result does not indicate that the first optimal solution R presented in (1) above should be adopted (step ST5; NO), the processing proceeds to step ST6.
[0047] In step ST6, the determination unit 18 determines whether the determination result indicates that the first optimal solution R is to be optimized based on the first aspiration level Q1 in (2) above (step ST6). If the determination result indicates that the first optimal solution R is to be optimized based on the first aspiration level Q1 in (2) above (step ST6; YES), the process proceeds to step ST7. On the other hand, if the determination result does not indicate that the first optimal solution R is to be optimized based on the first aspiration level Q1 in (2) above (step ST6; NO), the process proceeds to step ST8.
[0048] In step ST7, the first optimization calculation unit 14 searches for a solution that is in the vicinity of the first optimal solution R in the solution space K but is different from the first optimal solution R. The first optimization calculation unit 14 presents the searched solution to the decision maker via the interface unit 20, and the decision maker determines whether or not to adopt the presented solution. Thereafter, the first optimization calculation unit 14 repeats the search for the above solution until the presented solution is adopted by the decision maker. Note that, when the presented solution is adopted by the decision maker, the solution search device 1 ends the processing.
[0049] In step ST8, the determining unit 18 determines whether or not the determination result indicates that the first optimal solution R should be adjusted within the range of the Pareto frontier in (3) above (step ST8).
[0050] As a result, if it is determined in the above (3) that the judgment result does not indicate that the first optimal solution R should be adjusted within the range of the Pareto frontier (step ST8; NO), the aspiration level / ideal point setting unit 12 prompts the decision maker to re-input the first aspiration level Q1 via the interface unit 20. In response to this, the decision maker re-inputs the first aspiration level Q1 to the solution search device 1 via the interface unit 20. Thereafter, the processing returns to step ST3, where the aspiration level / ideal point setting unit 12 accepts the first aspiration level Q1 re-input by the decision maker via the interface unit 20 and sets the accepted first aspiration level Q1 in the solution space K. Thereafter, the solution search device 1 performs the processing from step ST3 onwards.
[0051] On the other hand, if it is determined that the above judgment result indicates that the first optimal solution R should be adjusted within the range of the Pareto frontier in (3) above (step ST8; YES), the processing proceeds to step ST9.
[0052] In step ST9, the first optimal solution adjustment unit 15 adjusts the first optimal solution R within the range of the Pareto frontier (step ST9). For example, the first optimal solution adjustment unit 15 moves the first optimal solution R on the line PF representing the frontier in the solution space K. At this time, the first optimal solution adjustment unit 15 receives an adjustment amount (movement amount) of the first optimal solution R from the decision maker and moves the first optimal solution R on the line PF representing the frontier based on the received adjustment amount. As a result, the first optimal solution adjustment unit 15 calculates a second optimal solution different from the first optimal solution R. The first optimal solution (i.e., the second optimal solution) adjusted as described above is presented to the decision maker via the interface unit 20.
[0053] Next, the second aspiration level calculation unit 16 calculates a second aspiration level different from the first aspiration level Q1 based on the first optimal solution (i.e., the second optimal solution) adjusted by the first optimal solution adjustment unit 15 and the ideal point P1 (step ST10). The calculated second aspiration level is presented to the decision maker via the interface unit 20.
[0054] An example of calculating the second optimal solution and the second aspiration level is shown in Figure 5. In Figure 5, point P1 indicates the ideal point, point Q1 indicates the first aspiration level, and point R indicates the first optimal solution. Also in Figure 5, point Q2 indicates the second aspiration level, and point R' indicates the second optimal solution.
[0055] 5 , the first optimal solution adjustment unit 15 adjusts the first optimal solution R by moving the first optimal solution R on the line PF indicating the Pareto frontier based on the adjustment amount received from the decision maker. In this way, the first optimal solution adjustment unit 15 calculates a second optimal solution R′ that is different from the first optimal solution R.
[0056] The second aspiration level calculation unit 16 calculates a second aspiration level Q2, which is an aspiration level different from the first aspiration level Q1, based on the second optimal solution R' calculated by the first optimal solution adjustment unit 15 and the ideal point P1. Details of calculation examples of the second optimal solution R' and the second aspiration level Q2 will be described later.
[0057] Next, the decision maker checks the presented second optimal solution R′, judges the above points (4) to (6), and inputs the judgment results to the solution search device 1 via the interface unit 20.
[0058] Next, the determination unit 18 determines whether the determination result indicates that the presented second optimal solution R' should be adopted in (4) above (step ST11). As a result, if it is determined that the determination result indicates that the presented second optimal solution R' should be adopted in (4) above (step ST11; YES), the process ends. On the other hand, if it is determined that the determination result does not indicate that the presented second optimal solution R' should be adopted in (4) above (step ST11; NO), the process proceeds to step ST12.
[0059] In step ST12, the judgment unit 18 judges whether the judgment result indicates that the second optimal solution R' is to be optimized based on the second aspiration level Q2 in the above (5) (step ST6). As a result, if it is judged that the judgment result indicates that the second optimal solution R' is to be optimized based on the second aspiration level Q2 in the above (5) (step ST12; YES), the processing proceeds to step ST13. On the other hand, if it is judged that the judgment result does not indicate that the second optimal solution R' is to be optimized based on the second aspiration level Q2 in the above (5) (step ST11; NO), the processing proceeds to step ST14.
[0060] In step ST13, the second optimization calculation unit 17 searches for a solution that is in the vicinity of the second optimal solution R' in the solution space K but is different from the second optimal solution R'. The second optimization calculation unit 17 presents the searched solution to the decision maker via the interface unit 20, and the decision maker determines whether or not to adopt the presented solution. Thereafter, the second optimization calculation unit 17 repeats the search for the above solution until the presented solution is adopted by the decision maker. Note that, when the presented solution is adopted by the decision maker, the solution search device 1 ends the processing.
[0061] In step ST14, the determination unit 18 determines whether or not the determination result indicates that the setting of the first aspiration level Q1 should be restarted in the above (6) (step ST14).
[0062] As a result, if it is determined that the above judgment result does not indicate that the setting of the first aspiration level Q1 should be restarted in the above (6) (step ST14; NO), the first optimal solution adjustment unit 15 prompts the decision maker to re-input the adjustment amount via the interface unit 20. In response to this, the decision maker re-inputs the adjustment amount to the solution search device 1 via the interface unit 20. Thereafter, the processing returns to step ST9, where the first optimal solution adjustment unit 15 accepts the adjustment amount re-input from the decision maker and moves the first optimal solution R on the line PF indicating the Pareto frontier based on the accepted adjustment amount. Thereafter, the solution search device 1 performs the processing from step ST9 onwards.
[0063] On the other hand, if it is determined in (6) above that the determination result indicates that the setting of the first aspiration level Q1 should be started over again (step ST14; YES), the aspiration level / ideal point setting unit 12 prompts the decision maker to re-input the first aspiration level Q1 via the interface unit 20. In response to this, the decision maker re-inputs the first aspiration level Q1 to the solution search device 1 via the interface unit 20. Thereafter, the processing returns to step ST3, where the aspiration level / ideal point setting unit 12 accepts the first aspiration level Q1 re-input by the decision maker via the interface unit 20 and sets the accepted first aspiration level Q1 in the solution space K. Thereafter, the solution search device 1 performs the processing from step ST3 onwards.
[0064] Next, a detailed calculation example of the second optimal solution R' and the second aspiration level Q2 will be described with reference to FIG.
[0065] In Fig. 6, the horizontal axis represents the first objective function (g1), and the vertical axis represents the second objective function (g2). Also in Fig. 6, point P1 represents the ideal point, point Q1 represents the first aspiration level, and point R represents the first optimal solution. Also in Fig. 6, point Q2 represents the second aspiration level, and point R' represents the second optimal solution.
[0066] 6, points C and D indicate solutions on the Pareto frontier. Here, points C and D are referred to as current solutions. In the example of FIG. 6, the first optimal solution R exists on the line segment connecting the current solution C and the current solution D.
[0067] 6, the coordinates of each point in the solution space K are defined as follows: First optimal solution R: (x, y) Second optimal solution R': (x', y') Ideal point P1: (x1, y1) First aspiration level Q1: (x0, y0) Second aspiration level Q2: (x'0, y'0) Current solution C: (x2, y2) Current solution D: (x3, y3)
[0068] Here, the first optimal solution adjustment unit 15 adjusts the first optimal solution R by moving the first optimal solution R on the line segment connecting the current solution C and the current solution D, and calculates a second optimal solution R' that is different from the first optimal solution R. At this time, the decision maker inputs the adjustment amount of the first optimal solution R as a coefficient t.
[0069] The coefficient t corresponds to the ratio between the length of the line segment connecting the current solution C and the current solution D in the solution space and the length of the line segment connecting the current solution C (or the current solution D) and the second optimal solution R', as shown in Fig. 6, for example. The decision maker inputs this coefficient t in the range of 0 ≤ t ≤ 1 via the interface unit 20. Then, the first optimal solution adjustment unit 15 uses the input coefficient t to calculate the second optimal solution R' according to the following equation (1).
[0070] For example, if the coefficient t is 0, the second optimal solution R' is equal to the current solution C, but as the coefficient t increases from 0, the second optimal solution R' approaches the current solution D on the line segment connecting the current solutions C and D. Then, when the coefficient t becomes 1, the second optimal solution R' becomes equal to the current solution D. Furthermore, when the coefficient t is set to 0.5, the second optimal solution R' is located at the midpoint between the current solutions C and D on the line segment connecting the current solutions C and D.
[0071] After the second optimal solution R' is calculated as described above, the second aspiration level calculation unit 16 calculates a second aspiration level Q2, which is an aspiration level different from the first aspiration level Q1, based on the calculated second optimal solution R' and the ideal point P1. For example, the second aspiration level calculation unit 16 calculates the second aspiration level Q2 using the following equation (2):
[0072] In equation (2), 1-s is the ratio of the length of the line segment connecting ideal point P1 and current solution D to the length of the line segment connecting ideal point P1 and point U. Point U can be uniquely determined as the intersection of a line connecting ideal point P1 and current solution D and a line passing through first aspiration level Q1 that is parallel to the line segment connecting current solution C and current solution D. In other words, the second aspiration level calculation unit 16 calculates the second aspiration level Q2 by applying the relationship of the ratio of the line segments that holds between ideal point P1, current solution D, and point U to the relationship between ideal point P1, the second optimal solution R', and the second aspiration level Q2.
[0073] In this manner, the first optimal solution adjustment unit 15 calculates the second optimal solution R' by adjusting the first optimal solution R within the range of the Pareto frontier based on the coefficient t, which is the adjustment amount specified by the decision maker. The second aspiration level calculation unit 16 calculates the second aspiration level Q2 based on the second optimal solution R' calculated by the first optimal solution adjustment unit 15 and the ideal point P1. In this case, the first optimal solution adjustment unit 15 calculates the second optimal solution R' by adjusting the first optimal solution R within the range of the Pareto frontier, rather than changing the aspiration level setting from the first aspiration level to the second aspiration level and obtaining the second optimal solution based on the changed second aspiration level, as in the conventional method. Therefore, the solution search device 1 can improve the search efficiency for the second optimal solution R' compared to the conventional method. Therefore, the solution search device 1 can shorten the time required to obtain an optimal solution that satisfies the decision maker.
[0074] Next, a detailed example of calculation of the first optimal solution R, the second optimal solution R', and the second aspiration level Q2 will be described with reference to FIGS. 7 and 8, taking an example in which the solution space K is a three-dimensional space.
[0075] 7 and 8, the solution space K is a three-dimensional space. That is, there are three objective functions. Here, as an example, the three objective functions are "age of building," "area," and "rent," and a multi-objective optimization problem is described in which the objective is to find a rental property that optimizes these three objective functions.
[0076] 7, point P1 indicates the ideal point, point Q1 indicates the first aspiration level, and point R indicates the first optimal solution. Also in FIG. 7, symbol PF indicates the surface representing the Pareto frontier, and points C to E indicate solutions (current solutions) on the Pareto frontier.
[0077] In addition, in FIG. 7, v1 indicates a vector from the ideal point P1 to the current solution C, v2 indicates a vector from the ideal point P1 to the current solution D, and v3 indicates a vector from the ideal point P1 to the current solution E. sol indicates a vector from the ideal point P1 to the first optimal solution R. In this case, V sol is also called the first vector, and v1 to v3 are also called the second vectors.
[0078] Even when the solution space K is a three-dimensional space, the solution search device 1 basically operates in the same way as when the solution space K is a two-dimensional space. For example, the first optimal solution calculation unit 13 calculates, in the solution space K, the intersection of a line connecting the ideal point P1 accepted by the aspiration level / ideal point setting unit 12 and the first aspiration level Q1 with a surface PF indicating the Pareto frontier as the first optimal solution R.
[0079] At this time, a vector V from the ideal point P1 to the first optimal solution R is sol can be expressed by the following equations (3) and (4) by adding v1, v2, and v3, where v1, v2, and v3 are assumed to be linearly independent.
[0080] In this case, equations (5) to (7) are obtained from equations (3) and (4).
[0081] Here, (α, β, γ) expressed in equation (7) corresponds to the adjustment amount (coefficient t) when the above-mentioned solution space K is a two-dimensional space. In other words, when the solution space K is a three-dimensional space, the decision maker inputs the adjustment amount of the first optimal solution R as the values of (α, β, γ). Then, the first optimal solution adjustment unit 15 adjusts the first optimal solution R on the plane PF using equation (3) based on the input (α, β, γ), and generates a new V sol , that is, the second optimal solution R' is calculated.
[0082] Similarly to the case where the solution space K is a two-dimensional space, the second aspiration level calculation unit 16 calculates a second aspiration level Q2, which is an aspiration level different from the first aspiration level Q1, based on the second optimal solution R' calculated by the first optimal solution adjustment unit 15 and the ideal point P1. For example, the second aspiration level calculation unit 16 calculates the second aspiration level Q2 using the following equation (8):
[0083] In other words, in the solution space K, the ratio between the distance from the ideal point P1 to the first aspiration level Q1 and the distance from the ideal point P1 to the first optimal solution R is determined, and the second aspiration level calculation unit 16 calculates the second aspiration level Q2 using this determined ratio and the distance from the ideal point P1 to the second optimal solution R'.
[0084] In this way, when the solution space K is a three-dimensional space, the first optimal solution adjustment unit 15 calculates the second optimal solution R' by adjusting the first optimal solution R within the range of the Pareto frontier based on the adjustment quantities (α, β, γ) specified by the decision maker. The second aspiration level calculation unit 16 calculates the second aspiration level Q2 based on the second optimal solution R' calculated by the first optimal solution adjustment unit 15 and the ideal point P1. Even in this case, the first optimal solution adjustment unit 15 calculates the second optimal solution R' by adjusting the first optimal solution R within the range of the Pareto frontier, rather than changing the aspiration level setting from the first aspiration level to the second aspiration level and obtaining the second optimal solution based on the changed second aspiration level, as in the conventional method. Therefore, the search efficiency for the second optimal solution R' is improved compared to the conventional method. Therefore, the time required to obtain an optimal solution that satisfies the decision maker can be shortened compared to the conventional method.
[0085] Next, an example of a hardware configuration of the solution search device 1 according to the first embodiment will be described with reference to Fig. 9 . The functions of the population generation unit 10, solution extraction unit 11, aspiration level / ideal point setting unit 12, first optimal solution calculation unit 13, first optimization calculation unit 14, first optimal solution adjustment unit 15, second aspiration level calculation unit 16, second optimization calculation unit 17, determination unit 18, display control unit 23, and screen generation unit 24 in the solution search device 1 are realized by processing circuits. The processing circuit may be dedicated hardware as shown in Fig. 9A , or may be a CPU (also referred to as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (digital signal processor)) 52 that executes a program stored in memory 53 as shown in Fig. 9B .
[0086] When the processing circuit is dedicated hardware, the processing circuit 51 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The functions of each of the population generation unit 10, the solution extraction unit 11, the aspiration level / ideal point setting unit 12, the first optimal solution calculation unit 13, the first optimization calculation unit 14, the first optimal solution adjustment unit 15, the second aspiration level calculation unit 16, the second optimization calculation unit 17, the determination unit 18, the display control unit 23, and the screen generation unit 24 may be realized individually by the processing circuit 51, or the functions of each unit may be realized collectively by the processing circuit 51.
[0087] When the processing circuit is a CPU 52, the functions of the population generation unit 10, solution extraction unit 11, aspiration level / ideal point setting unit 12, first optimal solution calculation unit 13, first optimization calculation unit 14, first optimal solution adjustment unit 15, second aspiration level calculation unit 16, second optimization calculation unit 17, judgment unit 18, display control unit 23, and screen generation unit 24 are realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in the memory 53. The processing circuit realizes the functions of each unit by reading and executing the programs recorded in the memory 53. In other words, the solution search device 1 includes a memory for storing a program that, when executed by the processing circuit, results in the execution of each step shown in FIG. 2, for example. In addition, these programs can also be said to cause a computer to execute the procedures and methods of the population generation unit 10, solution extraction unit 11, aspiration level / ideal point setting unit 12, first optimal solution calculation unit 13, first optimization calculation unit 14, first optimal solution adjustment unit 15, second aspiration level calculation unit 16, second optimization calculation unit 17, judgment unit 18, display control unit 23, and screen generation unit 24. Here, examples of the memory 53 include non-volatile or volatile semiconductor memories such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), and EEPROM (Electrically EPROM), magnetic disks, flexible disks, optical disks, compact disks, minidisks, and DVDs (Disc-like Versatile Discs).
[0088] Note that the functions of the population generation unit 10, the solution extraction unit 11, the aspiration level / ideal point setting unit 12, the first optimal solution calculation unit 13, the first optimization calculation unit 14, the first optimal solution adjustment unit 15, the second aspiration level calculation unit 16, the second optimization calculation unit 17, the judgment unit 18, the display control unit 23, and the screen generation unit 24 may be partially implemented by dedicated hardware and partially implemented by software or firmware. For example, the function of the population generation unit 10 may be implemented by a processing circuit as dedicated hardware, and the functions of the solution extraction unit 11, the aspiration level / ideal point setting unit 12, the first optimal solution calculation unit 13, the first optimization calculation unit 14, the first optimal solution adjustment unit 15, the second aspiration level calculation unit 16, the second optimization calculation unit 17, the judgment unit 18, the display control unit 23, and the screen generation unit 24 may be implemented by the processing circuit reading and executing programs stored in the memory 53.
[0089] Thus, the processing circuitry can implement each of the above-described functions by hardware, software, firmware, or a combination thereof.
[0090] In the above description, an example has been described in which the solution search device 1 includes a population generation unit 10 and a solution extraction unit 11, and generates a population (a set of solutions) and extracts a frontier using the population generation unit 10 and the solution extraction unit 11. However, the solution search device 1 is not limited to this, and may acquire information about a frontier extracted in advance by an external device from the device and perform processing using the acquired information about the frontier. In this case, the population generation unit 10 and the solution extraction unit 11 may be omitted.
[0091] In the above description, an example has been described in which the aspiration level / ideal point setting unit 12 receives an ideal point and a first aspiration level input by a decision maker via the interface unit 20 and sets the received ideal point and first aspiration level in the solution space. However, the aspiration level / ideal point setting unit 12 does not necessarily have to receive an ideal point and a first aspiration level from a decision maker. For example, the aspiration level / ideal point setting unit 12 may read an ideal point and a first aspiration level that have been set in advance in the solution search device 1 and set them in the solution space.
[0092] Furthermore, in the solution search device 1, the processes of optimizing the first optimal solution R based on the first aspiration level Q1 (step ST7 in FIG. 2), calculating the second aspiration level Q2 (step ST10 in FIG. 2), and optimizing the second optimal solution R' based on the second aspiration level Q2 (step ST13 in FIG. 2) do not necessarily need to be performed. Furthermore, if these processes are omitted, the first optimization calculation unit 14, the second aspiration level calculation unit 16, and the second optimization calculation unit 17 are not essential components of the solution search device 1 and may be omitted.
[0093] As described above, according to the first embodiment, the solution search device 1 includes the aspiration level / ideal point setting unit 12 that sets the first aspiration level Q1, which is a target value for multiple objective functions in a multi-objective optimization problem, the first optimal solution calculation unit 13 that calculates the first optimal solution R using the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12 and a frontier, which is a solution set for optimizing the multiple objective functions, and the first optimal solution adjustment unit 15 that, when receiving an instruction from the user not to adopt the first optimal solution R calculated by the first optimal solution calculation unit 13, receives an input of an adjustment amount for adjusting the first optimal solution R from the user and adjusts the first optimal solution R within the range of the frontier based on the received adjustment amount. As a result, the solution search device 1 according to the first embodiment can shorten the time required to obtain an optimal solution that satisfies a decision maker compared to conventional methods.
[0094] Furthermore, the first optimal solution adjustment unit 15 adjusts the first optimal solution R within the range of the frontier based on the adjustment amount, thereby calculating a second optimal solution R' that is different from the first optimal solution R. This allows the solution search device 1 according to the first embodiment to accurately calculate the second optimal solution R'.
[0095] Furthermore, when the solution space of the multi-objective optimization problem is a two-dimensional space, the first optimal solution calculation unit 13 calculates, as the first optimal solution R, the intersection of a line connecting the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12 and the ideal point P1 that indicates the optimal solution for all of the objective functions, and a line indicating the frontier in the solution space, and the first optimal solution adjustment unit 15 adjusts the first optimal solution R calculated by the first optimal solution calculation unit 13 by the adjustment amount on the line indicating the frontier. As a result, the solution search device 1 according to the first embodiment can accurately calculate the first optimal solution R and the second optimal solution R' when the solution space is a two-dimensional space.
[0096] Furthermore, when the first optimal solution R is located between two current solutions that are solutions existing on the line indicating the frontier, the adjustment amount is given by the ratio of the length of the line segment connecting the two current solutions to the length of the line segment connecting one of the two current solutions with the adjusted first optimal solution R'. This allows the solution search device 1 according to the first embodiment to accurately calculate the second optimal solution R' based on the adjustment amount.
[0097] Furthermore, when the solution space of the multi-objective optimization problem is a three-dimensional space, the first optimal solution calculation unit 13 calculates the intersection of a line connecting the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12 and an ideal point P1 that indicates an optimal solution for all of the objective functions among the plurality of objective functions, and a plane that indicates a frontier in the solution space, as the first optimal solution R, and the first optimal solution adjustment unit 15 adjusts the first optimal solution R calculated by the first optimal solution calculation unit 13 by moving it on the plane that indicates the frontier by the adjustment amount. As a result, the solution search device 1 according to the first embodiment can accurately calculate the first optimal solution R and the second optimal solution R' when the solution space is a three-dimensional space.
[0098] Furthermore, when the first optimal solution R is located on a plane that represents the frontier and that includes the three current solutions that are solutions included in the frontier, the adjustment amount is given by a coefficient by which each of the second vectors is multiplied when the first vector leading from the ideal point P1 to the adjusted first optimal solution R' is expressed as the sum of three second vectors leading from the ideal point P1 to each of the three current solutions. This allows the solution search device 1 according to the first embodiment to accurately calculate the second optimal solution R' based on the adjustment amount.
[0099] The solution search device 1 also includes a second aspiration level calculation unit 16 that calculates a second aspiration level Q2 that is an aspiration level different from the first aspiration level Q1, based on the first optimal solution R' adjusted by the first optimal solution adjustment unit 15 and an ideal point P1 that indicates the optimal solution for all of the objective functions among the plurality of objective functions. This allows the solution search device 1 according to the first embodiment to accurately calculate the second aspiration level Q2 that is an aspiration level different from the first aspiration level Q1.
[0100] Embodiment 2 In the first embodiment, a solution search device 1 was described that calculates a second optimal solution R' that is different from the first optimal solution R by adjusting the first optimal solution R within the range of the frontier based on an adjustment amount received from a decision maker. In the second embodiment, a solution search device 1 having a GUI function for supporting the decision maker in inputting the adjustment amount will be described.
[0101] The configuration example of the solution search device 1 according to the second embodiment is the same as the configuration example of the solution search device 1 according to the first embodiment shown in Fig. 1. In the second embodiment, the screen generator 24 generates data (hereinafter also referred to as "screen data") showing a support screen for supporting the decision maker in inputting adjustment amounts. The display controller 23 causes the display unit 22 to display the support screen shown by the screen data generated by the screen generator 24.
[0102] An example of the support screen is shown in Fig. 10. Here, as an example, a case will be described in which the multi-objective optimization problem has two objective functions: rent and building age.
[0103] As shown in Figure 10, the support screen is composed of an ideal point input area A1, a first optimal solution display area A2, an adjustment amount input area A3, a solution space display area A4, and a current solution display area A5.
[0104] The ideal point etc. input area A1 is an area for the decision maker to input the ideal point P1 and the first aspiration level Q1 to the solution search device 1. In the ideal point etc. input area A1, the item "Rent_hope" indicates the first aspiration level for rent, and the item "Age of building_hope" indicates the first aspiration level for age of building. In addition, the item "Rent_ideal" indicates the ideal point for rent, and the item "Age of building_ideal" indicates the ideal point for age of building. In the example of FIG. 10 , the first aspiration level for rent is "50,000 (yen)", the first aspiration level for age of building is "10 (years)", the ideal point for rent is "28,000.00 (yen)", and the ideal point for age of building is "1.00 (years)".
[0105] In addition, an adjustment section (here, a slide button) for adjusting the ideal point P1 and the first aspiration level Q1 is displayed adjacent to each item. By appropriately operating this adjustment section, the decision maker can arbitrarily adjust the ideal point P1 and the first aspiration level Q1 while continuously changing the ideal point P1 and the first aspiration level Q1.
[0106] The first optimum solution display area A2 is an area where the first optimum solution R (shown as "virtual solution" in the figure) calculated by the first optimum solution calculation unit 13 is displayed.
[0107] The adjustment amount input area A3 is an area for the decision maker to input the adjustment amount, i.e., the coefficient t. In the example of Fig. 10, the coefficient t is set to "0.70", but this value may be set in advance in the solution search device 1 as an initial value.
[0108] Also, an adjustment section (here, a slide button) for adjusting the coefficient t is displayed adjacent to this item. By appropriately operating this adjustment section, the decision maker can arbitrarily adjust the coefficient t while continuously changing the value of the coefficient t. Note that the adjustment amount input area A3 also displays the adjusted optimal solution (i.e., the second optimal solution R') and the adjusted desired solution (i.e., the second desired level Q2).
[0109] The solution space display area A4 is an area for displaying the solution space. In the example of FIG. 10 , a two-dimensional solution space defined by the building age on the horizontal axis and the rent on the vertical axis is displayed. The solution space displayed in the solution space display area A4 displays the Pareto frontier extracted by the solution extraction unit 11, the solutions on the Pareto frontier, the ideal point P1, the first aspiration level Q1, the second aspiration level Q2, the first optimal solution R, and the second optimal solution R'. The decision maker can intuitively grasp each of the above values through this solution space display area A4.
[0110] The current solution display area A5 is an area for displaying the current solutions. Here, among the solutions that make up the Pareto frontier, two solutions (current solutions 1 and 2) that are particularly close to the first aspiration level Q1 are displayed as the current solutions.
[0111] Next, an example of transition of the support screen will be described. First, when the solution extraction unit 11 extracts the Pareto frontier, the screen generation unit 24 generates screen data in which the Pareto frontier is displayed on the solution space in the solution space display area A4. The display control unit 23 causes the display unit 22 to display the support screen indicated by the generated screen data.
[0112] Next, the decision maker inputs an ideal point P1 and a first aspiration level Q1 through the ideal point etc. input area A1 of the displayed support screen. In response to this input, the first optimal solution calculation unit 13 calculates a first optimal solution R. At this time, the screen generation unit 24 updates the previously generated screen data. The updated screen data is screen data in which the first optimal solution R is displayed in the first optimal solution display area A2, the first optimal solution R is displayed on the solution space in the solution space display area A4, and the current solution is displayed in the current solution display area A5. The display control unit 23 then displays the support screen indicated by the updated screen data on the display unit 22.
[0113] Next, the decision maker determines whether to adopt the first optimal solution R. If the decision is not to adopt the first optimal solution R and not to optimize the first optimal solution R based on the first aspiration level Q1, the decision maker inputs a coefficient t as an adjustment amount in the adjustment amount input area A3.
[0114] In response to this input, the first optimal solution adjustment unit 15 adjusts the first optimal solution R within the range of the Pareto frontier based on the coefficient t. As a result, the first optimal solution adjustment unit 15 calculates a second optimal solution R'. Furthermore, the second aspiration level calculation unit 16 calculates a second aspiration level Q2.
[0115] At this time, the screen generation unit 24 re-updates the previously updated screen data. The re-updated screen data is screen data in which the adjusted optimal solution (i.e., the second optimal solution R') and the adjusted aspiration solution (i.e., the second aspiration level Q2) are displayed in the adjustment amount input area A3, and the second optimal solution R' and the second aspiration level Q2 are displayed on the solution space in the solution space display area A4. The display control unit 23 causes the display unit 22 to display the support screen indicated by the re-updated screen data.
[0116] In this way, in the second embodiment, a support screen for supporting the decision maker in inputting the adjustment amount can be generated and displayed. This allows the decision maker to easily input the adjustment amount, reducing the operational burden. Furthermore, the decision maker can easily input the ideal point P1 and the first aspiration level Q1 via the support screen.
[0117] Note that the configuration example of the support screen shown in Figure 10 is merely one example. For example, the layout of the above-mentioned areas A1 to A5 may be other than those described above. Also, in the above explanation, an example was described in which the adjustment units displayed in the ideal point etc. input area A1 and the adjustment amount input area A3 were configured with slide buttons, but the adjustment units are not limited to this and may be configured with elements other than slide buttons.
[0118] As described above, according to the second embodiment, the solution search device 1 includes the screen generator 24 that generates data showing a support screen for supporting the decision maker in inputting adjustment amounts, and the display controller 23 that causes the display unit 22 to display the support screen shown by the data generated by the screen generator 24. As a result, the solution search device 1 according to the second embodiment can reduce the operational burden on the decision maker in addition to the effects of the first embodiment.
[0119] The support screen also includes an adjustment amount input area A3 in which the decision maker can input an adjustment amount and in which the first optimal solution R' adjusted by the first optimal solution adjustment unit 15 based on the input adjustment amount can be displayed. This allows the solution search device 1 according to the second embodiment to reduce the burden on the decision maker of inputting the adjustment amount. Furthermore, the decision maker can easily check the adjusted first optimal solution R'.
[0120] The support screen also includes an ideal point etc. input area A1 in which the decision maker can input a first aspiration level Q1 and an ideal point P1 that indicates an optimal solution for all of the objective functions among the plurality of objective functions. This allows the solution search device 1 according to the second embodiment to reduce the burden on the decision maker of inputting the ideal point P1 and the first aspiration level Q1.
[0121] Third Embodiment In the first embodiment, the case where the frontier is the Pareto frontier has been described as an example. In the third embodiment, the case where the frontier is the efficient frontier will be described as an example.
[0122] The configuration example of the solution search device 1 according to the third embodiment is the same as the configuration example of the solution search device 1 according to the first embodiment shown in Fig. 1. In the third embodiment, the solution extraction unit 11 extracts the efficient frontier using Data Envelopment Analysis (DEA) instead of the Pareto frontier.
[0123] Data envelopment analysis (DEA) is a tool for measuring the management efficiency of a business entity (DMU: Decision Making Unit) with a multi-input, multi-output system. It assumes that the business entity's production activities involve some input (input) and some revenue (output), and evaluates the conversion efficiency using a one-dimensional scale called the efficiency value.
[0124] For example, a case where efficiency values of eight business entities (A to J) having two objective functions g1 and g2 are calculated will be described with reference to Fig. 11. In Fig. 11, the horizontal axis represents the first objective function (g1) and the vertical axis represents the second objective function (g2).
[0125] In Figure 11, the further to the bottom left (the origin) an entity is, the more productive its activities are. In Data Envelopment Analysis (DEA), the efficient frontier in Figure 11 is considered to be the Pareto optimal solution, and each entity is evaluated by measuring how far it is from this efficient frontier.
[0126] In this case, entities on the frontier are judged to be conducting efficient activities and are given an evaluation value of 1 (=100%). Conversely, entities not on the frontier are judged to be conducting inefficient activities and are given an evaluation value of less than 1. In this case, the standard for evaluating each entity is given as a frontier rather than a single point (any one Pareto solution), making it possible to evaluate entities based on their characteristics.
[0127] When extracting the efficient frontier, the solution extraction unit 11 calculates the efficiency value of each solution (each business entity in the above example) existing in the solution space. At this time, examples of the method for calculating the efficiency value include the Chames-Cooper-Rhodes (CCR) model and the Ultra-CCR model.
[0128] The CCR model is a method for quantitatively evaluating efficiency values, proposed by Chames, Cooper, and Rhodes in 1978. This method is well known, so a detailed explanation will be omitted. In this method, the upper limit of the efficiency value is set to 1, and the efficient frontier is found.
[0129] For example, in the example of Fig. 12, the efficiency value θ of entity H is calculated as θ = OP / OH. Then, entities that maximize this θ (θ = 1) are extracted based on the following equation (9), and the efficient frontier is found from the extraction results. In equation (9), x represents input, y represents output, and u and v represent weights.
[0130] On the other hand, the Ultra CCR model is a model in which each business entity is removed from the constraints of the CCR model, and efficiency values are allowed to exceed 1. The frontier surface that serves as the basis for evaluation collapses, but business entities that were on the frontier in the CCR model can now calculate efficiency values of 1 or higher.
[0131] For example, in the example of Fig. 13, the efficiency value θ of business entity D is calculated as θ = OP / OD. Then, based on the following formula (10), business entities that maximize this θ (θ may be 1 or greater) are extracted, and the efficient frontier is found from the extraction results.
[0132] Here, the CCR (Chames Cooper Rhodes) model and the ultra-CCR model have been described as methods for calculating efficiency values, but the methods for calculating efficiency values are not limited to these, and other methods may also be used.
[0133] As described above, the solution extraction unit 11 in the third embodiment extracts an efficient frontier using data envelopment analysis (DEA) instead of the Pareto frontier. As a result, in the third embodiment, solutions with an efficiency value of 1 or more can be extracted as the frontier, and the number of solutions to be used as the frontier can be reduced compared to the first embodiment in which the Pareto frontier is extracted. Furthermore, in the third embodiment, since the number of solutions to be used as the frontier can be reduced compared to the first embodiment, interpretation of the solutions becomes easier.
[0134] As described above, according to the third embodiment, the solution search device 1 includes the solution extraction unit 11 that uses data envelope analysis to extract an efficient frontier as a frontier from a set of solutions that exist in the solution space of the multi-objective optimization problem, and the first optimal solution calculation unit 13 calculates the first optimal solution R using the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12 and the efficient frontier extracted by the solution extraction unit 11. As a result, the solution search device 1 according to the third embodiment can reduce the number of solutions to be used as a frontier compared to the first embodiment, in addition to the effects of the first embodiment.
[0135] Embodiment 4 In the first embodiment, an example was described in which a first optimal solution R is adjusted within the range of the frontier based on an adjustment amount received from a decision maker, thereby calculating a second optimal solution R' different from the first optimal solution R. In the fourth embodiment, an example is described in which, when the decision maker is not satisfied with the second optimal solution R' obtained by the adjustment, a new aspiration level (second aspiration level) is presented to the decision maker, and the second optimal solution R' is calculated from this new aspiration level.
[0136] Fig. 14 is a diagram showing an example of the configuration of a solution search device 1 according to embodiment 4. The solution search device 1 according to embodiment 4 additionally includes an aspiration level frontier calculation unit 31, a first aspiration level adjustment unit 32, and a second optimal solution calculation unit 33, in addition to the solution search device 1 according to embodiment 1 shown in Fig. 1. The other components of the solution search device 1 according to embodiment 4 are the same as those of the solution search device 1 according to embodiment 1 shown in Fig. 1, and therefore the same reference numerals are used and their description will be omitted.
[0137] When the aspiration level frontier calculation unit 31 receives an instruction from the decision maker not to adopt the first optimal solution (i.e., the second optimal solution) R' adjusted by the first optimal solution adjustment unit 15, it uses the above frontier and the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12 to calculate an aspiration level frontier, which is a set of aspiration level candidates including the first aspiration level Q1.
[0138] 15 shows an example of the aspiration level frontier calculated by the aspiration level frontier calculation unit 31. When the solution space of the multi-objective optimization problem is a two-dimensional space, as shown in FIG. 15, for example, the aspiration level frontier calculation unit 31 calculates the aspiration level frontier indicated by a line obtained by enlarging the line indicating the Pareto frontier by a predetermined magnification.
[0139] For example, the aspiration level frontier calculation unit 31 draws a line that passes through the first aspiration level Q1 and that represents the Pareto frontier, projecting it in the right-back direction (the direction of the first aspiration level Q1) in the solution space while enlarging it at a predetermined magnification. The line QF drawn in this way represents the aspiration level frontier. In this case, the curvature of the bent points on the line QF is approximately equal to the curvature of the bent points on the line representing the Pareto frontier.
[0140] The first aspiration level adjustment unit 32 receives input of an adjustment amount for adjusting the first aspiration level Q1 from the decision maker, and calculates the second aspiration level Q2 by adjusting the first aspiration level Q1 within the range of the aspiration level frontier based on the received adjustment amount.
[0141] The second optimal solution calculation unit 33 calculates a second optimal solution R' that is different from the first optimal solution R using the second aspiration level Q2 calculated by the first aspiration level adjustment unit 32 and the above-mentioned frontier.
[0142] Next, a calculation example of the second optimal solution R' and the second aspiration level Q2 in the fourth embodiment will be described with reference to FIG. 16. In FIG. 16, the horizontal axis represents the first objective function (g1), and the vertical axis represents the second objective function (g2). Also in FIG. 16, point P1 represents the ideal point, point Q1 represents the first aspiration level, and point R represents the first optimal solution. Also in FIG. 16, point Q2 represents the second aspiration level, and point R' represents the second optimal solution.
[0143] 16, points C and D indicate solutions (current solutions) on the Pareto frontier. In the example of FIG. 16, the first optimal solution R exists on the line segment connecting the current solution C and the current solution D. In addition, in FIG. 16, the line segment connecting points Q1 and Q2 is part of the line indicating the aspiration level frontier calculated by the aspiration level frontier calculation unit 31.
[0144] 16, the coordinates of each point in the solution space K are defined as follows: First optimal solution R: (x, y) Second optimal solution R': (x', y') Ideal point P1: (x1, y1) First aspiration level Q1: (x0, y0) Second aspiration level Q2: (x'0, y'0) Current solution C: (x2, y2) Current solution D: (x3, y3)
[0145] Here, the first aspiration level adjustment unit 32 adjusts the first aspiration level Q1 by moving the first aspiration level Q1 on the line segment representing the aspiration level frontier, and calculates a second aspiration level Q2 that is different from the first aspiration level Q1. At this time, the decision maker inputs, via the interface unit 20, one of the values (x, y) of the two objective functions that determine the second aspiration level Q2, as the amount of adjustment of the first aspiration level Q1.
[0146] For example, the decision maker inputs the value of the x-coordinate as an adjustment amount out of the x-coordinate and y-coordinate that define the second aspiration level Q2. In this case, the slope of the line segment connecting current solution C and current solution D is equal to the slope of the line segment connecting point Q1 and point Q2 (both are (y3-y2) / (x3-x2)). Therefore, if the decision maker simply inputs the value of the x-coordinate of the second aspiration level Q2, the first aspiration level adjustment unit 32 can calculate the value of the y-coordinate of the second aspiration level Q2 using the following formula (11).
[0147] Alternatively, the decision maker may input the value of the y-coordinate of the second aspiration level Q2 as the adjustment amount. In this case, the first aspiration level adjustment unit 32 can calculate the x-coordinate of the second aspiration level Q2 using the following equation (12).
[0148] In this way, the decision maker can allow the first aspiration level adjustment unit 32 to calculate the second aspiration level Q2 simply by inputting at least one of the x and y coordinates that define the second aspiration level Q2, i.e., the value of at least one of the two objective functions, as an adjustment amount, thereby reducing the burden of input operations on the decision maker.
[0149] Then, when the second aspiration level Q2 is calculated as described above, the second optimum solution calculation unit 33 calculates the second optimum solution R' using the following equation (13).
[0150] In equation (13), 1-s is the ratio of the length of the line segment connecting the ideal point P1 and the current solution D to the length of the line segment connecting the ideal point P1 and point U. Point U can be uniquely determined as the intersection of the line segment connecting the ideal point P1 and the current solution D and the line segment representing the aspiration level frontier. In other words, the second optimal solution calculation unit 33 calculates the second optimal solution R' by applying the relationship of the ratio of the line segments that holds between the ideal point P1, the current solution D, and point U to the relationship between the ideal point P1, the second optimal solution R', and the second aspiration level Q2.
[0151] As described above, in the fourth embodiment, an aspiration level frontier is generated, and the first aspiration level Q1 is adjusted within the range of this aspiration level frontier to calculate the second aspiration level Q2. The second optimal solution R' is then calculated from this second aspiration level Q2. This means that the second optimal solution R' is calculated using a different approach from that of the first embodiment. As a result, even if a satisfactory solution cannot be obtained by adjusting the first optimal solution R within the range of the frontier, the decision maker can obtain the second optimal solution R' using a different approach, i.e., by adjusting the first aspiration level Q1 within the range of the aspiration level frontier. Furthermore, in this case, the decision maker only needs to input at least one of the x- and y-coordinates that define the second aspiration level Q2 as the adjustment amount for the first aspiration level Q1, thereby reducing the input operation load.
[0152] In the above explanation, it is assumed that the decision maker adjusts the first optimal solution R within the range of the frontier, but this is not limited to this, and the decision maker may generate the aspiration level frontier and adjust the first aspiration level Q1 without adjusting the first optimal solution R within the range of the frontier.
[0153] As described above, according to the fourth embodiment, the solution search device 1 includes an aspiration level frontier calculation unit 31 that, when receiving an instruction from a decision maker not to adopt the first optimal solution R′ adjusted by the first optimal solution adjustment unit 15, calculates an aspiration level frontier, which is a set of aspiration level candidates including the first aspiration level Q1, using the frontier and the first aspiration level Q1 set by the aspiration level / ideal point setting unit 12; a first aspiration level adjustment unit 32 that receives an input of an adjustment amount for adjusting the first aspiration level Q1 from the decision maker and calculates a second aspiration level Q2 by adjusting the first aspiration level Q1 within the range of the aspiration level frontier based on the received adjustment amount; and a second optimal solution calculation unit 33 that calculates a second optimal solution R′ different from the first optimal solution R, using the second aspiration level Q2 calculated by the first aspiration level adjustment unit 32 and the frontier. As a result, in addition to the effects of embodiment 1, the solution search device 1 according to embodiment 4 can obtain the second aspiration level Q2 by adjusting the first aspiration level Q1. Furthermore, a new optimal solution R' can be calculated based on the obtained second aspiration level Q2.
[0154] Furthermore, when the solution space of the multi-objective optimization problem is a two-dimensional space, the aspiration level frontier calculation unit 31 calculates the aspiration level frontier indicated by the line QF obtained by enlarging the line indicating the frontier by a predetermined magnification. This allows the solution search device 1 according to the fourth embodiment to easily generate the aspiration level frontier.
[0155] Furthermore, the adjustment amount is given by the value of one of the two objective functions that determine the second aspiration level Q2. This allows the solution search device 1 according to the fourth embodiment to reduce the burden of input operations on the decision maker.
[0156] Fifth Embodiment In the fifth embodiment, a specific method of optimization calculation by the first optimization calculation unit 14 and the second optimization calculation unit 17 will be described. The configuration example of the solution search device 1 according to the fifth embodiment is the same as the configuration example of the solution search device 1 according to the first embodiment shown in FIG.
[0157] When a decision maker determines to optimize the first optimal solution R based on the first aspiration level Q1, the first optimization calculation unit 14 searches for solutions that are located around the first optimal solution R in the solution space K and that are different from the first optimal solution R. At this time, the first optimization calculation unit 14 converts the multi-objective optimization problem into a single-objective optimization problem based on the ideal point P1, the first aspiration level Q1, and the first optimal solution R, thereby determining a search region for solutions and performing an intensive search for solutions in the search region.
[0158] Similarly, when the decision maker determines to optimize the second optimal solution R' based on the second aspiration level Q2, the second optimization calculation unit 17 searches for a solution different from the second optimal solution R' that is located around the second optimal solution R' in the solution space K. At this time, the second optimization calculation unit 17 converts the multi-objective optimization problem into a single-objective optimization problem based on the ideal point P1, the second aspiration level Q2, and the second optimal solution R', thereby determining a search area for a solution and conducting an intensive search for a solution in the search area.
[0159] The basic solution search methods used by the first optimization calculation unit 14 and the second optimization calculation unit 17 are similar, so the basic solution search method used by the second optimization calculation unit 17 will be described here as an example.
[0160] First, the second optimization calculation unit 17 sets a region in the solution space K that includes the second optimal solution R' (hereinafter also referred to as the "reference region"), and sets a region in the solution space K that is an extension of this reference region to a predetermined range (hereinafter also referred to as the "extended region").The second optimization calculation unit 17 then randomly samples multiple solutions included in this extended region and calculates an evaluation value for each of the obtained solutions.
[0161] An example of this case is shown in Fig. 17. For example, in Fig. 17, symbol R' is the second optimal solution, symbol K1 is the reference region, and symbol K2 is the extended region. Symbol gi is a sampled solution included in the extended region K2.
[0162] The second optimization calculation unit 17 calculates the evaluation value Vn,i of the sampled solution g i using, for example, the following equation (14).
[0163] That is, the second optimization calculation unit 17 calculates the evaluation value Vn,i of each solution g i based on how far each solution g i is from the second aspiration level Q2 in the solution space K. Here, the farther a solution g i is from the second aspiration level Q2 (the closer a solution g i is to the ideal point P1), the higher the evaluation value Vn,i that is calculated.
[0164] The second optimization calculation unit 17 then selects a predetermined number of solutions g i with the highest evaluation values V n,i. Then, as shown in Fig. 18, the second optimization calculation unit 17 sets a predetermined region in the solution space that includes each of the selected solutions g i as a new search region K3, and randomly samples multiple solutions g i' included in this search region K3. The solutions g i' sampled from the search region K3 are then presented to the decision maker via the interface unit 20, and the decision maker determines whether or not to adopt each of the presented solutions g i'.
[0165] As described above, in the fifth embodiment, the second optimization calculation unit 17 converts the multi-objective optimization problem into a single-objective optimization problem by calculating the evaluation values Vn,i based on the ideal point P1 and the second aspiration level Q2 for the solutions g i sampled from the extended region K2 including the second optimal solution R'. The second optimization calculation unit 17 then selects a predetermined number of solutions g i with the highest evaluation values Vn,i, sets a predetermined region in the solution space including each of the selected solutions g i as a new search region K3, and focuses the search on the solutions g i' included in this search region K3. This allows the fifth embodiment to focus the search on solutions close to the decision maker's preferences.
[0166] The solution search method by the first optimization calculation unit 14 is basically the same as the solution search method by the above-mentioned second optimization calculation unit 17. When using the first optimization calculation unit 14, in the above explanation, the second optimal solution R' may be read as the first optimal solution R, and the second aspiration level Q2 may be read as the first aspiration level Q1.
[0167] As described above, according to the fifth embodiment, the solution search device 1 includes the second optimization calculation unit 17 that searches for solutions existing in the peripheral region of the adjusted first optimal solution R' in the solution space K of the multi-objective optimization problem when the solution search device 1 receives an instruction from the decision maker not to adopt the first optimal solution R' adjusted by the first optimal solution adjustment unit 15. In this way, the solution search device 1 according to the fifth embodiment can, in addition to the effects of the first embodiment, focus on searching for solutions that are close to the decision maker's preferences.
[0168] Furthermore, the second optimization calculation unit 17 calculates evaluation values of solutions existing in the area surrounding the adjusted first optimal solution R' based on the second aspiration level Q2 and the ideal point P1, and further searches around a predetermined number of solutions with the highest calculated evaluation values. In this way, the solution search device 1 according to the fifth embodiment can convert the multi-objective optimization problem into a single-objective optimization problem using the evaluation values, and can focus on searching for solutions that are close to the decision maker's preferences.
[0169] Sixth Embodiment In a sixth embodiment, another example of a specific technique for optimization calculation by the first optimization calculation unit 14 and the second optimization calculation unit 17 will be described. The configuration example of the solution search device 1 according to the sixth embodiment is the same as the configuration example of the solution search device 1 according to the first embodiment shown in FIG.
[0170] When a decision maker determines to optimize the first optimal solution R based on the first aspiration level Q1, the first optimization calculation unit 14 searches for solutions that are located around the first optimal solution R in the solution space K and are different from the first optimal solution R. At this time, the first optimization calculation unit 14 determines a search area for solutions using the first aspiration level Q1 as a reference point and performs an intensive search for solutions in that search area.
[0171] Similarly, when the decision maker determines to optimize the second optimal solution R' based on the second aspiration level Q2, the second optimization calculation unit 17 searches for solutions different from the second optimal solution R' that are located around the second optimal solution R' in the solution space K. At this time, the second optimization calculation unit 17 determines a search area for solutions using the second aspiration level Q2 as a reference point, and performs an intensive search for solutions in that search area.
[0172] The basic solution search methods used by the first optimization calculation unit 14 and the second optimization calculation unit 17 are similar, so the basic solution search method used by the second optimization calculation unit 17 will be described here as an example.
[0173] First, the second optimization calculation unit 17 sets a second aspiration level Q2 as a reference point in the solution space. Then, the second optimization calculation unit 17 sets a region in the solution space that can include the region (reference region) containing the second optimal solution R' and that is within a predetermined distance from the reference point as a new search region K4, and focuses on searching for solutions g' included in this search region K4. The searched solutions g' are then presented to the decision maker via the interface unit 20, and the decision maker determines whether or not to adopt each presented solution g'.
[0174] As described above, in the sixth embodiment, the second optimization calculation unit 17 sets the second aspiration level Q2 as a reference point in the solution space, sets a search area K4 based on the distance from this reference point, and focuses on searching for solutions g i′ included in this search area K4. This makes it possible to focus on searching for solutions close to the decision maker's preferences in the sixth embodiment.
[0175] The solution search method by the first optimization calculation unit 14 is basically the same as the solution search method by the above-mentioned second optimization calculation unit 17. When using the first optimization calculation unit 14, in the above explanation, the second optimal solution R' may be read as the first optimal solution R, and the second aspiration level Q2 may be read as the first aspiration level Q1.
[0176] As described above, according to the sixth embodiment, the second optimization calculation unit 17 uses the second aspiration level Q2 as a reference point and searches for an area in the surrounding area of the adjusted first optimal solution R' whose distance from the reference point in the solution space K is within a predetermined value. As a result, in the sixth embodiment, in addition to the effects of the first embodiment, it is possible to focus on searching for solutions that are close to the decision maker's preferences.
[0177] Finally, an application example of the solution search device 1 according to the present disclosure will be described. The solution search device 1 according to the present disclosure can be applied to, for example, an air conditioning optimization system.
[0178] For example, a building manager may change the temperature setting of an air conditioning system based on past power consumption data to manage daily power consumption so as not to exceed the target value. However, if the manager suppresses power consumption, there is a risk that the comfort (thermal comfort) of the office may decrease. Therefore, the manager sets an appropriate desired level for the air conditioning optimization system, taking into account thermal comfort and allowable power consumption.
[0179] On the other hand, an air conditioning optimization system calculates an optimal solution based on an aspiration level set by a manager in a multi-objective optimization problem in which thermal comfort is a first objective function and power consumption is a second objective function, and presents the solution to the manager. The solution search device 1 according to the present disclosure can be applied to air conditioning optimization systems used in such situations.
[0180] Furthermore, the solution search device 1 according to the present disclosure can be applied to, for example, a surveillance camera placement optimization system.
[0181] For example, surveillance cameras (hereinafter also referred to as "cameras") are installed everywhere, such as in commercial facilities, train stations, and city streets. Generally, cameras are installed to detect signs of crimes, accidents, fights, and other incidents to prevent them before they occur, or to check whether an incident has occurred and grasp the situation at the scene. Therefore, cameras must be positioned so that they can reliably capture the situation, such as what happened when an incident occurs or what is currently happening. However, increasing the number of cameras not only increases costs, but also risks increasing the burden of maintaining and managing the cameras.
[0182] Therefore, the administrator of the camera placement optimization system takes into consideration the areas that need to be monitored with priority (for example, areas that are prone to blind spots) and the number of cameras that are acceptable, and sets an appropriate level of preference for the camera placement optimization system.
[0183] On the other hand, a camera placement optimization system calculates an optimal solution based on an aspiration level set by an administrator in a multi-objective optimization problem in which the camera monitoring range is a first objective function and the number of cameras is a second objective function, and presents the solution to the administrator. The solution search device 1 according to the present disclosure can be applied to a camera placement optimization system used in such a situation.
[0184] In addition, the present disclosure allows for free combination of the embodiments, modification of any of the components of the embodiments, or omission of any of the components of the embodiments.
[0185] The present disclosure makes it possible to shorten the time required to obtain an optimal solution that satisfies a decision maker, and is suitable for use in a solution search device and a solution search method.
[0186] 1 Solution search device, 10 Population generation unit, 11 Solution extraction unit, 12 Aspiration level / ideal point setting unit, 13 First optimal solution calculation unit, 14 First optimization calculation unit, 15 First optimal solution adjustment unit, 16 Second aspiration level calculation unit, 17 Second optimization calculation unit, 18 Judgment unit, 20 Interface unit, 21 Input unit, 22 Display unit, 23 Display control unit, 24 Screen generation unit, 31 Aspiration level frontier calculation unit, 32 First aspiration level adjustment unit, 33 Second optimal solution calculation unit, 51 Processing circuit, 52 CPU, 53 Memory, A1 Ideal point etc. input area, A2 First optimal solution display area, A3 Adjustment amount input area, A4 Solution space display area, A5 Solution display area, C Current solution, D Current solution, g1 Objective function, g2 Objective function, gi Solution, gi' Solution, K Solution space, K2 expansion region, K3 search region, K4 search region, P1 ideal point, PF line or surface representing the frontier, Q1 first aspiration level, Q2 second aspiration level, R first optimal solution, R' second optimal solution.
Claims
1. A solution search device comprising: a setting unit that sets a first aspiration level, which is a target value for multiple objective functions in a multi-objective optimization problem; a first optimal solution calculation unit that calculates a first optimal solution using the first aspiration level set by the setting unit and a frontier, which is a solution set for optimizing the multiple objective functions; and a first optimal solution adjustment unit that, when receiving an instruction from a user not to adopt the first optimal solution calculated by the first optimal solution calculation unit, receives an input of an adjustment amount for adjusting the first optimal solution from the user, and adjusts the first optimal solution within the range of the frontier based on the received adjustment amount.
2. The solution search device according to claim 1, characterized in that the first optimal solution adjustment unit calculates a second optimal solution different from the first optimal solution by adjusting the first optimal solution within the range of the frontier based on the adjustment amount.
3. The solution search device according to claim 1 or 2, characterized in that when a solution space for the multi-objective optimization problem is a two-dimensional space, the first optimal solution calculation unit calculates, as a first optimal solution, an intersection point between a line connecting the first aspiration level set by the setting unit and an ideal point indicating optimal solutions for all of the multiple objective functions, and a line indicating the frontier in the solution space, and the first optimal solution adjustment unit performs the adjustment by moving the first optimal solution calculated by the first optimal solution calculation unit by the adjustment amount on the line indicating the frontier.
4. A solution search device as described in claim 3, characterized in that, when the first optimal solution is located between two current solutions which are solutions existing on the line indicating the frontier, the adjustment amount is given by the ratio between the length of the line segment connecting the two current solutions and the length of the line segment connecting either of the two current solutions with the first optimal solution after the adjustment.
5. A solution search device as claimed in any one of claims 1 to 4, characterized in that when a solution space for the multi-objective optimization problem is a three-dimensional space, the first optimal solution calculation unit calculates, as a first optimal solution, an intersection between a line connecting the first aspiration level set by the setting unit and an ideal point indicating optimal solutions for all of the objective functions among the plurality of objective functions, and a plane indicating the frontier in the solution space, and the first optimal solution adjustment unit performs the adjustment by moving the first optimal solution calculated by the first optimal solution calculation unit by the adjustment amount on the plane indicating the frontier.
6. A solution search device as described in claim 5, characterized in that when the first optimal solution is located on a plane representing the frontier and including three current solutions which are solutions included in the frontier, the adjustment amount is given by a coefficient to be multiplied by each of the second vectors when the first vector leading from the ideal point to the first optimal solution after the adjustment is expressed as the sum of three second vectors leading from the ideal point to each of the three current solutions.
7. A solution search device as described in any one of claims 1 to 6, characterized in that it comprises a second aspiration level calculation unit that calculates a second aspiration level, which is an aspiration level different from the first aspiration level, based on the first optimal solution adjusted by the first optimal solution adjustment unit and an ideal point that indicates the optimal solution in all of the multiple objective functions.
8. A solution search device as described in any one of claims 1 to 7, characterized in that it comprises: a screen generation unit that generates data showing an assistance screen for assisting the user in inputting the adjustment amount; and a display control unit that displays on a display unit the assistance screen shown by the data generated by the screen generation unit.
9. The solution search device according to claim 8, characterized in that the support screen includes an area in which the user can input the adjustment amount and in which a first optimal solution adjusted by the first optimal solution adjustment unit based on the input adjustment amount can be displayed.
10. The solution search device described in claim 8 or claim 9, characterized in that the support screen includes an area in which the user can input the first aspiration level and an ideal point indicating an optimal solution for all of the multiple objective functions.
11. A solution search device as claimed in any one of claims 1 to 10, characterized in that it comprises a solution extraction unit that uses data envelopment analysis to extract an efficient frontier as the frontier from a set of solutions existing in a solution space for the multi-objective optimization problem, and the first optimal solution calculation unit calculates a first optimal solution using a first aspiration level set by the setting unit and the efficient frontier extracted by the solution extraction unit.
12. A solution search device as described in any one of claims 1 to 11, characterized in that it comprises: an aspiration level frontier calculation unit that, when receiving an instruction from a user not to adopt the first optimal solution adjusted by the first optimal solution adjustment unit, calculates an aspiration level frontier, which is a set of aspiration level candidates including the first aspiration level, using the frontier and the first aspiration level set by the setting unit; a first aspiration level adjustment unit that receives an input of an adjustment amount for adjusting the first aspiration level from the user and calculates a second aspiration level by adjusting the first aspiration level within the range of the aspiration level frontier based on the received adjustment amount; and a second optimal solution calculation unit that calculates a second optimal solution different from the first optimal solution, using the second aspiration level calculated by the first aspiration level adjustment unit and the frontier.
13. The solution search device according to claim 12, characterized in that, when the solution space in the multi-objective optimization problem is a two-dimensional space, the aspiration level frontier calculation unit calculates the aspiration level frontier represented by a line obtained by enlarging the line representing the frontier by a predetermined magnification.
14. The solution search device according to claim 13, characterized in that the adjustment amount is given by the value of one of two objective functions that define the second aspiration level.
15. The solution search device according to claim 7, further comprising an optimization calculation unit that, when receiving an instruction from a user not to adopt the first optimal solution adjusted by the first optimal solution adjustment unit, searches for solutions existing in a peripheral area of the adjusted first optimal solution in a solution space for the multi-objective optimization problem.
16. The solution search device according to claim 15, characterized in that the optimization calculation unit calculates an evaluation value of a solution existing in a surrounding area of the adjusted first optimal solution based on the second aspiration level and an ideal point indicating an optimal solution in all of the objective functions among the plurality of objective functions, and further searches the areas surrounding a predetermined number of solutions having the highest calculated evaluation values.
17. The solution search device according to claim 15, characterized in that the optimization calculation unit uses the second aspiration level as a reference point and searches for an area surrounding the adjusted first optimal solution, the distance in solution space from the reference point being within a predetermined value.
18. A solution search method by a solution search device, comprising: a setting unit setting a first aspiration level, which is a target value for multiple objective functions in a multi-objective optimization problem; a first optimal solution calculation unit calculating a first optimal solution using the first aspiration level set by the setting unit and a frontier, which is a solution set for optimizing the multiple objective functions; and a first optimal solution adjustment unit, when receiving an instruction from a user not to adopt the first optimal solution calculated by the first optimal solution calculation unit, receiving an input of an adjustment amount for adjusting the first optimal solution from the user, and adjusting the first optimal solution within the range of the frontier based on the received adjustment amount.
Citation Information
Patent Citations
Method and system for optimizing operation of power generating plant
JP2009030476A
System and method of interactive, multi-objective visualization
US7921371B1
