JP5007664B2 - Optimal design calculation apparatus, optimal design calculation program and method - Google Patents

Description

  The present invention relates to an optimum design technique based on computer simulation.

  In recent years, optimal design using computer simulation has been actively performed in order to shorten the design and development period against the backdrop of low price and high performance of computers and the spread of simulation software and high accuracy.

  Among computer simulations, Monte Carlo simulation is used in a wide range of applications such as communication device design and business risk analysis, and can be said to be one of the particularly important simulation methods.

  In Monte Carlo simulation, for example, various operating environments of a device are set using random numbers, whether or not the device performs an expected operation under each operating condition, and device performance such as an abnormal operation rate is calculated. be able to.

  In the optimum design, device performance is calculated for design examples corresponding to various values of device design parameters, and design parameter values corresponding to better device performance are searched. When optimal design is performed using Monte Carlo simulation, various design examples are set by Monte Carlo simulation, device performance is calculated for each design example, and processing for comparing and evaluating device performance values is performed.

  By the way, regarding the relationship between the calculation accuracy and the calculation amount of Monte Carlo simulation, in order to increase the calculation accuracy by 10 times, it is necessary to increase the calculation amount by trial of random numbers by 100 times. For this reason, the Monte Carlo simulation has a relatively high calculation load.

  With regard to optimal design technology, for example, threshold values corresponding to a plurality of types of parameters related to equipment design in mobile communications are stored, and the evaluation values for each parameter are compared by comparing the obtained design values with the threshold values. There is provided a technique for determining the above (for example, Patent Document 1).

  Alternatively, as an analysis technique in the optimum design, assembly position data (explanatory variable) measured for each of a plurality of parts and components, and reference assembly position data (object variable) indicating the assembly accuracy of the final finished product or intermediate finished product A technique for performing regression analysis based on the above and identifying a component having a high contribution rate with respect to assembly accuracy is provided (for example, Patent Document 2).

  As an analysis technique, there is provided a technique for calculating an evaluation index of a response waveform of a PID control parameter, performing fuzzy inference based on the calculated evaluation index, and adjusting a PID parameter based on the fuzzy inference result (for example, Patent Document 3).

  As a technique for improving search accuracy in optimal design, a technique for generating a plurality of response surfaces corresponding to a plurality of combinations of design parameters and parameters representing response characteristics and searching for design parameters satisfying predetermined response characteristics is provided. (For example, Patent Document 4).

On the other hand, data mining is one method for analyzing a large amount of data and searching for correlations and similar relationships (for example, Non-Patent Documents 1, 2, and 3).
JP 2003-114911 A JP 2006-264521 A JP-A-6-131007 JP 2003-16114 A Atsushi Otaki et al., “Applied binary tree analysis method”, Nikka Giren Publisher, July 1998 Michael J. A. Berry et al., "Data Mining Technique", Kaibundo Publishing, September 1999 W. N. Venables et al., “Statistical analysis with S-PLUS”, Springer Fairlake Tokyo, July 2001.

  In the past, designers have performed optimal design by adjusting calculation accuracy and calculation amount based on their own experience. For this reason, there has been a problem that the result of optimum design depends on the experience of the designer. In addition, there is a problem that optimization work cannot be performed in a time zone when the designer is absent.

  Further, in the conventional optimum design calculation system, the Monte Carlo simulation and the calculation of the optimum design are performed using different software. For this reason, conventionally, adjustment of the calculation time of the Monte Carlo simulation has been performed without referring to the value of the apparatus performance obtained in the calculation of the optimum design.

  The present invention provides a technology that can make efficient use of computer resources by enabling high-speed calculation of optimal design and enabling optimal design regardless of the skill level of the user. The purpose is to provide.

In order to solve the above problems, a computer that performs an optimal design calculation by performing a Monte Carlo simulation for a design example that is a combination of design parameters, and a Monte Carlo simulation for a plurality of design examples are performed, and the operation of each device is normal. in it and to give a false trial number operation is determined to be abnormal in the determined true number of attempts and the device, device performance value calculating means for calculating an equipment performance values from said vacuum number of trials and該偽number of trials The device performance value obtained by performing the Monte Carlo simulation on the i-th design example of the plurality of design examples by the device performance calculation means is obtained by performing the Monte Carlo simulation on the design example up to the i−1th design example. Probability values that are better than the best device performance values are assigned to the best device performance values and i-th design examples of the i-1th design examples. If the probability value calculation means calculated using the original binomial distribution and the true number of trials and false trial number, the calculated probability values in the probability value calculating means is smaller than the predetermined threshold, the device performance When the Monte Carlo simulation by the value calculation means is interrupted and the probability value is equal to or greater than a predetermined threshold value, an optimum design calculation program that functions as a determination means for continuing the Monte Carlo simulation by the apparatus performance value calculation means is provided.

  A probability value is obtained that the device performance value calculated by performing the i-th Monte Carlo simulation for a design example is better than the best device performance value among the device performance values calculated up to the (i-1) th. When the obtained probability value is less than the predetermined threshold value, the Monte Carlo simulation for the design example is interrupted. By using the result of the Monte Carlo simulation for the calculation of the optimum design, the speed of the optimum design calculation can be increased. In addition, if a probability value obtained for a design example is less than the expected threshold value, the Monte Carlo simulation for that design example is interrupted, regardless of the skill level of the user. In addition, it is possible to perform optimal design, and it is possible to effectively use computer resources.

  The probability value may be calculated using the number of true trials and the number of false trials for design examples other than the design example in which the i th Monte Carlo simulation is performed. For example, design examples included in the same region may be taken out using regression tree analysis and used for calculation of probability values. Design examples included in the same group may be taken out by cluster analysis and used to calculate probability values. A similar design example may be taken out by inference analysis based on memory and used for calculation of probability values.

  Furthermore, based on the device performance value obtained by performing the Monte Carlo simulation with a predetermined number of trials on a certain design example, the computer can obtain a design parameter range corresponding to the device performance value below the predetermined value by data mining. Design parameter range calculation means to be calculated, a change width of the device performance value corresponding to the design parameter range calculated by the design parameter range calculation means, and a Monte Carlo simulation to be performed on another design example from the calculated change width A trial number setting means for setting the number of trials, whether or not a Monte Carlo simulation has been performed for a predetermined number of design examples, and a Monte Carlo simulation is performed for the predetermined number of design examples Ending the Monte Carlo simulation by the device performance value calculating means Determining means may be configured to further function as a.

  A design parameter range in which a good device performance value can be obtained is obtained, and when a Monte Carlo simulation is performed for the obtained design parameter range, the number of trials is set so that the device performance value can be calculated with high accuracy. In the initial stage of calculation, since the design parameter range is set to be wide, the number of trials is set to be small, so that the calculation process can be speeded up. At the end of the calculation, the design parameter range is narrowed, but since the number of trials is set to a large value, a highly accurate apparatus performance value can be obtained. Thereby, it is possible to speed up the calculation processing as a whole and improve the reliability of the required device performance value.

  According to the optimal design calculation program disclosed above, the Monte Carlo simulation and the optimal design can be integrated to speed up the calculation of the optimal design. Furthermore, optimal design can be performed regardless of the level of proficiency of the user, thereby making it possible to effectively use computer resources.

DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings.
<First Embodiment>
FIG. 1 is a configuration diagram of an optimum design calculation system according to the first embodiment. An optimum design calculation system 1 shown in FIG. 1 includes a management computer 2 and a calculation computer 3, and the computers are connected to each other via a network. In this embodiment, the management computer and the calculation computer are constituted by personal computers (PCs).

  In this embodiment, a plurality of calculation personal computers (hereinafter referred to as calculation PCs) 3 are provided, and each calculation PC 3 (calculation PCs 3A, 3B, and 3C in FIG. 1) stores a Monte Carlo simulation program from the memory. Read out and execute Monte Carlo simulation. The optimum design by Monte Carlo simulation will be described later.

  The management personal computer (hereinafter referred to as management PC) 2 reads the optimization / statistical analysis calculation program from the memory, calculates the device performance value from the result of the simulation received from the calculation PC 3, and calculates the calculated device performance value. Optimize using.

  FIG. 2 is a block diagram of the optimum design calculation system 1 according to the present embodiment. As described above, in the optimum design calculation system 1, a plurality of calculation PCs 3 are provided, but since the configuration of each calculation PC 3 is the same, only one calculation PC 3 is shown in FIG. ing. Below, the outline | summary of the optimal design process in the optimal design calculation system 1 which concerns on this embodiment is demonstrated.

  First, the management PC 2 gives an execution start instruction to start the Monte Carlo simulation to the calculation PC 3 according to the information input from the user, and gives the design parameters obtained by the optimization program to the calculation PC 3. . Details of the design parameters will be described with reference to FIG.

  The calculation PC 3 executes the Monte Carlo simulation program using the given design parameters in accordance with the simulation execution start instruction input from the management PC 2. Specifically, the device performance value is calculated from the number of true trials and the number of false trials obtained by executing the simulation. The number of true trials, the number of false trials, and the device performance value are passed from the calculation PC 3 to the management PC 2.

  The management PC 2 analyzes using the statistical analysis program, and if it is determined that a sufficiently good device performance value is obtained, the management PC 2 instructs the calculation PC 3 to interrupt execution of the Monte Carlo simulation. The determination of the device performance value here will be described later. When it is determined that a device performance value equal to or less than a predetermined value has not been obtained, the Monte Carlo simulation is continued using the given design parameter. When the number of simulation trials reaches a predetermined number, the design parameter End the simulation. When the simulation is interrupted or terminated, the management PC 2 outputs the simulation result to a data file.

  The above processing is repeated while changing the design parameter value. The management PC 2 records the number of true trials, the number of false trials, and the device performance value for each design parameter in the data file. The data file will be described with reference to FIG.

FIG. 3 is a diagram for explaining the optimum design by Monte Carlo simulation.
According to the optimal design by Monte Carlo simulation, for example, when performing optimal design for a certain device and verifying the operation of the device under various operating conditions, whether or not the device operates normally with respect to the operating conditions generated using random numbers. To simulate. Search for operating conditions that can minimize the abnormal operating rate and improve the reliability of the device.

  As illustrated in FIG. 3, when a random number value is generated, a Monte Carlo simulation is performed to determine normality / abnormality of the operation of the apparatus under operation conditions such as a data amount corresponding to the random value and noise. Then, the abnormal operation rate is obtained as a device performance value. In the example of FIG. 3, as a result of performing the Monte Carlo simulation four times, the normal operation is three times and the abnormal operation is one time. In this case, the abnormal operation rate, that is, the device performance value is obtained as 1/4 = 25%.

  4A is a diagram illustrating an example of a design parameter search range, and FIG. 4B is a diagram illustrating an example of the design parameter set in accordance with the search range of FIG. 4A. Here, the case where a design example is comprised from three design parameters X, Y, and Z is mentioned as an example.

As shown in FIG. 4A, a lower limit value and an upper limit value are set for each set parameter in the design parameter search range. In the embodiment, the lower limit “10” and the upper limit “40” are set for the design parameter X. For the design parameter Y, the lower limit value “20” and the upper limit value “50” are set, and for the design parameter Z, the lower limit value “30” and the upper limit value “60” are set.

  The management PC refers to the design parameter search range shown in FIG. 4A, calculates design parameters included in the range, and passes the obtained values to the calculation PC. When the calculation PC receives a combination of design parameters as shown in FIG. 4B, the calculation PC executes a simulation of the combination of design parameters, that is, a design example.

FIG. 5 is a diagram illustrating an example of a data file. The device performance value, the number of true trials, and the number of false trials are recorded for each design example, that is, for each combination of design parameters.
Of the data recorded in the data file, the number of true trials is the number of trials determined as a result of the simulation that the operation of the device is normal, and the number of false trials is the abnormal operation of the device The number of trials determined as. The device performance value is a value calculated from the number of true trials and the number of false trials, and is defined by the following equation.
(Device performance value) = (Number of false trials) / (Number of true trials + Number of false trials) (1)
Calculation is performed by substituting the result obtained by performing the Monte Carlo simulation for the i-th (i is a natural number) for a certain design example into the above equation (1). Then, using the binomial distribution, the probability that the i-th calculated device performance value is better than the best value among the device performance values calculated up to the (i−1) -th for the design example is calculated. To do. Specifically, it is obtained from the following equation (2).

In the equation (2), NT represents the number of true trials, NF represents the number of false trials, and P _best represents the best, that is, the smallest apparatus performance value among the apparatus performance values calculated up to the (i-1) th.
In this embodiment, the probability value calculated by the above equation (2) is compared with the threshold value for the probability set in advance, and the probability value calculated from the above equation (2) is the threshold value. If it is less than that, the Monte Carlo simulation is interrupted. When the probability value calculated from the above equation (2) is equal to or greater than the threshold value, the Monte Carlo simulation is continued, but the number of trials for the design example performing the simulation reaches a preset upper limit value. Ends the process.

  When the Monte Carlo simulation is interrupted or terminated for one design example, the result is recorded in a data file. An optimization program is executed based on the data recorded in the data file, and a design example to be simulated next is obtained. Thereafter, the above process is repeated for a predetermined number of design examples.

FIG. 6 is a flowchart showing the optimum design calculation process according to the first embodiment.
First, in step S1, when the user inputs the upper limit value of the number of iterations, the search range of design parameters, the upper limit value of the number of trials, and the threshold value of the probability of interruption into the optimum design calculation system 1, in step S2, A new data file is created in the management PC 2 and a counter for the number of simulation trials i (i is a natural number) for a design example and a counter for the number of iterations j indicating the number of design examples for simulation are initially set. Turn into. In step S3, the management PC 2 executes an optimization program, and calculates and outputs a design parameter value estimated to be optimal based on the input design parameter search range and data file.

  Note that the number of design examples to be calculated is set as the upper limit value of the number of repeated calculations. As the upper limit value of the number of trials, an upper limit value of the number of times the Monte Carlo simulation is performed for a certain design example is set. As the threshold value of the probability of interruption, the probability value represented by equation (2) when a value that is expected to be sufficiently good is obtained as the device performance value of the device that is the target of the optimum design is set. The

  In step S4, the calculation PC 3 initializes by setting “0” as the value of the number of true trials and the number of false trials. In step S5, the calculation PC 3 sets an operation condition of the apparatus using a random number, and performs a Monte Carlo simulation on the operation of the apparatus. Then, it is determined whether the operation is normal or abnormal. If the operation is determined to be normal in step S6, 1 is added to the number of true trials, and if the operation is determined to be abnormal, 1 is added to the number of false trials.

  In step S7, the probability value that the device performance value for the design example that has been simulated this time in the PC 3 for calculation is better than the device performance value that takes the best value among the device performance values that have been obtained by simulation until the previous time. Is calculated based on the number of true trials and the number of false trials. The probability value is obtained by the above equation (2). Then, it is determined whether or not the obtained probability value and the number of simulation trials satisfy a predetermined condition.

  That is, when (i) the probability value is less than the above threshold value of the probability of interruption, or (ii) the sum of the number of true trials and the number of false trials is equal to the upper limit of the number of trials Then, the Monte Carlo simulation is terminated, and the process proceeds to step S8. If neither of the above conditions (i) and (ii) is satisfied, the counter for the number of trials i is incremented by 1 and the process returns to step S5. For the same design example, the Monte Carlo simulation from step S5 to step S7 is performed. Go to calculate the number of true trials / number of false trials and repeat the process of finding the probability value.

  In step S8, the performance value of the apparatus is calculated from the above-described equation (1) in the calculation PC3. In step S9, the data received from the calculation PC3 for the design example in which the current Monte Carlo simulation is performed in the management PC2, that is, The design parameter, the device performance value, the number of true trials and the number of false trials are additionally output to the data file. In step S10, the management PC 2 determines whether or not the number of iterations is equal to the upper limit of the number of iterations. If the number of iterations is less than the upper limit, the counter for the number of iterations j is set to 1. Add and return to step S3. Then, referring to the data file, design parameters estimated to be optimal by the optimization program are set, and the same processing is executed for the set design example. If the number of repeated calculations becomes equal to the upper limit value in step S10, the process is terminated.

  As described above, according to the optimum design calculation system according to the present embodiment, a preset interruption probability threshold value is set, and an i-th apparatus performance value calculated for a design example is i. The probability value that is better than the best value among the device performance values calculated up to −1 is calculated using a binomial distribution and compared with a threshold value of interruption probability. When the obtained probability value is smaller than the interruption probability threshold value, the optimal design calculation process is interrupted because a sufficiently good apparatus performance value can be obtained in the design example in which the i th Monte Carlo simulation is performed.

  When the obtained probability value is equal to or greater than the threshold value of the probability of interruption, the Monte Carlo simulation is continued until the probability value becomes less than the threshold value of the probability of interruption. However, an upper limit is set in advance for the number of times the Monte Carlo simulation is performed for a certain design example, and when the number of simulation trials reaches the upper limit, the process ends for that design example.

  In this way, the Monte Carlo simulation is interrupted by comparing the obtained probability value with the threshold value of the interruption probability, and the simulation is terminated when the number of trials reaches the upper limit value, thereby calculating the Monte Carlo simulation. This prevents the time required for the device performance value calculation process from being prolonged while maintaining high accuracy. Thereby, even an unskilled user can easily and optimally design. In addition, by making it possible to perform optimal design regardless of the skill level of the user, it becomes possible to effectively use computer resources.

  Furthermore, when the calculation of the device performance value by Monte Carlo simulation for a design example is interrupted or terminated, optimization is performed with reference to the data file that records the simulation results, and the design parameters (design examples) to be simulated next are determined. calculate. By adopting a configuration that uses device performance values obtained by Monte Carlo simulation for optimization, it is possible to speed up the processing compared to the conventional calculation and optimization of device performance values. It becomes.

<Second Embodiment>
An optimum design calculation system 1 according to the second embodiment will be described. In the first embodiment, the probability value is obtained for each design example, whereas in the optimal design calculation system according to the second embodiment, the same area as the design example in which the Monte Carlo simulation is performed by regression tree analysis is used. The design example to which the design belongs is extracted and the probability value is calculated using the number of true trials and the number of false trials for the design examples in the same region.

The configuration diagram and the block diagram of the optimum design calculation system 1 according to the present embodiment are the same as those according to the first embodiment, and thus the description thereof is omitted here.
FIG. 7 is a diagram for explaining regression tree analysis. According to regression tree analysis, the correspondence between input values (input parameters) and output values (output evaluation indicators) is analyzed, and threshold values are set so that input parameters are classified according to representative values of output evaluation indicators. The input parameter value is calculated. For regression tree analysis, for example, known techniques described in Non-Patent Documents 1 and 2 above are used.

  A graph 40 shows the relationship between the input parameter and the representative value of the output evaluation index when the region is divided. In the graph 40, the output evaluation index is divided into two regions using the value “10” of the variable A as a threshold value, and the values “5” and “10” of the variable B for each of the divided regions. Is divided into two areas. The relationship between the input parameters (variable A (VarA) and variable B (VarB)) shown in the graph 40 and the representative value of the output evaluation index is represented by a rule 50.

  That is, regarding the relationship between the representative value of the output evaluation index and the variables A and B, the representative value of the output evaluation index is 0.2 for the region where the variable A is less than 10 and the variable B is less than 5. For the region where the variable A is less than 10 and the variable B is 5 or more, the representative value of the output evaluation index is 0.5, and for the region where the variable A is 10 or more and the variable B is less than 15, the representative of the output evaluation index When the value is 0.7, the variable A is 10 or more, and the variable B is 15 or more, the representative value of the output evaluation index is 0.9.

  When a tree is generated according to the rule 50, a regression tree 30 shown in FIG. 7 is obtained. The regression tree 30 is branched depending on whether the value of the variable A is less than 10. If the variable A is less than 10, the process further branches depending on whether the value of the variable B is less than 5. If the variable A is 10 or more, the process branches depending on whether the variable B is less than 15.

  Thus, generating the regression tree 30 branched according to the threshold value of the input parameter is synonymous with obtaining the rule 50 and dividing the region according to the value of the input parameter. In this embodiment, a rule representing the relationship between the design parameter corresponding to the input parameter and the apparatus performance value corresponding to the output evaluation index is obtained by regression tree analysis.

  FIG. 8 is a diagram illustrating a relationship between design parameters obtained by regression tree analysis and apparatus performance values. Each of the points shown in the graph of FIG. 8A represents a design example, and the coordinates of the points indicating the design example represent the values of two design parameters X and Y, respectively.

  In the example shown in FIG. 8A, from the relationship between the design example and the apparatus performance value, an area where a good apparatus performance value can be obtained, an area where an intermediate apparatus performance value can be obtained, and a low apparatus performance value can be obtained. As a result of searching for a threshold value to divide into regions, the region is divided into four regions, that is, regions A, B, C, and C. Of the four areas, the area A is an area (optimum area) including a design example in which a good apparatus performance value can be obtained.

  FIG. 8B is a diagram illustrating an example of a regression tree analysis rule obtained as a result of the regression tree analysis. The region is divided into regions A and B and regions C and D by the threshold value “20” for the design parameter X. Regions A and B are divided into two by the threshold value “30” for the design parameter Y, and regions C and D are divided into two by the threshold value “40” for the design parameter Y. .

  In the present embodiment, as described above, the design example is divided into regions by regression tree analysis, and the number of true trials and the number of false trials for the region to which the design example to which the Monte Carlo simulation is performed belongs is simulated. Are added to the number of true trials and the number of false trials, respectively, and the probability value is calculated based on the total number of true trials and false trials obtained.

FIG. 9 is a flowchart showing the optimum design calculation process according to the second embodiment. A description will be given centering on differences from the optimum design calculation processing according to the first embodiment shown in FIG.
Steps S21 to S26 correspond to steps S1 to S6 in FIG. 6, respectively. When the number of true trials and the number of false trials for the design example for which the Monte Carlo simulation has been performed is obtained in step S26, the process proceeds to step S27.

If there is one or more design example data recorded in the data file in step S27, regression tree analysis is performed on the data file. The regression tree analysis is as described with reference to FIG. Then, another design example belonging to the same area as the design example that has been simulated in step S25 is extracted from the data file, and the number of true trials and the number of false trials are obtained for the extracted design example. The formulas for obtaining the number of true trials and the number of false trials in step S27 are as shown in the following formulas (3) and (4).
(Number of true trials) = (Number of true trials obtained in step S26) + (Total number of true trials for extracted design examples) (3)
(Number of false trials) = (Number of false trials obtained in step S26) + (Total number of false trials for extracted design examples) (4)
The processing after step S28 is the same as the processing after step 7 in FIG.

As described above, according to the optimum design calculation system according to this embodiment, design examples belonging to the same region as the design example in which the i-th apparatus performance value is calculated using regression tree analysis are extracted. Using the number of true trials and false trials for the extracted design example and the number of true trials and false trials for the design example for which the simulation was performed, the i-th calculated device performance value is up to i-1 The probability of obtaining a value better than the best value among the device performance values calculated in (1) is calculated. By finding the probability value including the number of true trials and the number of false trials for design examples belonging to the same area by regression tree analysis, not only can the optimal design process be accelerated, but also the accuracy of the probability value can be improved. It becomes possible to calculate the apparatus performance value more appropriately.

<Third Embodiment>
An optimum design calculation system 1 according to the third embodiment will be described. In the second embodiment, the region is divided by regression tree analysis, whereas in the present embodiment, the design example is grouped by cluster analysis.

The configuration diagram and the block diagram of the optimum design calculation system 1 according to the present embodiment are the same as those according to the first embodiment, and thus the description thereof is omitted here.
FIG. 10 is a diagram for explaining cluster analysis. As in FIG. 8A, each of the points shown in the graph of FIG. 10 represents a design example, and the coordinates of the points indicating the design example represent the values of the design parameters X and Y. For the cluster analysis, for example, a known technique described in Non-Patent Document 2 is used.

  In the present embodiment, cluster analysis is used to classify design examples (design parameters) that are similar to each other, that is, in which design parameter values are close to each other, into one group. In the example shown in FIG. 10, it is classified into four groups of groups A, B, C, and D.

  In this embodiment, design examples are grouped by cluster analysis, and the number of true trials and the number of false trials for the design examples included in the same group as the design example for which the Monte Carlo simulation was performed are calculated. The above probability value is calculated based on the total number of true trials and false trials obtained by adding to the number of trials and false trials.

  FIG. 11 is a flowchart showing the optimum design calculation process according to the third embodiment. Description will be made mainly on differences from the optimum design calculation process using regression tree analysis, that is, the optimum design calculation process according to the second embodiment shown in FIG.

  Steps S41 to S46 correspond to steps S21 to S26 in FIG. 9, respectively. When the number of true trials and the number of false trials for the design example for which the Monte Carlo simulation has been performed are obtained in step S46, the process proceeds to step S47.

  If there is one or more design example data recorded in the data file in step S47, cluster analysis is performed on the data file. The cluster analysis is as described with reference to FIG. Then, other design example data belonging to the same group as the design example that has been simulated in step S45 is extracted from the data file, and the number of true trials and the number of false trials are obtained for the extracted design examples. The number of true trials and the number of false trials are obtained from the above formulas (3) and (4).

The processing after step S48 is the same as the processing after step S28 in FIG.
As described above, according to the optimum design calculation system according to the present embodiment, design examples in the same group as the design example in which the i-th apparatus performance value is calculated using cluster analysis are extracted. Using the number of true trials and false trials for the extracted design example and the number of true trials and false trials for the design example for which the simulation was performed, the i-th calculated device performance value is up to the (i-1) th. The probability of obtaining a value better than the best value among the device performance values calculated in (1) is calculated. By calculating the probability value including the number of true trials and false trials for the design example of the same group by cluster analysis, not only can the optimal design process be accelerated, but the accuracy of the probability value is further improved. Therefore, it is possible to calculate the apparatus performance value more appropriately.

<Fourth Embodiment>
An optimum design calculation system 1 according to the fourth embodiment will be described. In the second and third embodiments described above, design examples are extracted using regression tree analysis and cluster analysis, respectively, whereas in this embodiment, simulation is performed by inference analysis based on memory. The difference is that a design example similar to the above-described design example is searched and the design example obtained by the search is used.

The configuration diagram and the block diagram of the optimum design calculation system according to the present embodiment are the same as those according to the first embodiment, and thus the description thereof is omitted here.
FIG. 12 is a diagram for explaining inference analysis based on memory. Similarly to FIG. 8A and FIG. 10, each point on the graph in the figure represents a design example, and the coordinates of the points indicating the design example represent the values of the design parameters X and Y. For the reasoning analysis based on memory, for example, a known technique described in Non-Patent Document 2 is used.

  A design example similar to the design example for which the i-th apparatus performance value is calculated is retrieved and extracted. The number of true trials and the number of false trials for the extracted design example are added to the number of true trials and the number of false trials for the design example for which the i-th apparatus performance value was calculated by simulation, respectively, and the number of true trials obtained And the above probability value based on the total number of false trials.

  FIG. 13 is a flowchart showing the optimum design calculation process according to the fourth embodiment. Description will be made mainly on differences from the above-described optimal design calculation processing using regression tree analysis and cluster analysis, that is, the optimal design calculation processing according to the second and third shown in FIGS.

  Steps S61 to S66 correspond to steps S21 to S26 in FIG. 9 and steps S41 to S46 in FIG. 11, respectively. When the number of true trials and the number of false trials for the design example for which the Monte Carlo simulation has been performed are obtained in step S66, the process proceeds to step S67.

  If there is one or more design example data recorded in the data file in step S67, inference analysis based on storage is performed for the data file. The inference analysis based on the memory is as described with reference to FIG. Then, design example data similar to the design example simulated in step S65 this time is extracted from the data file, and the number of true trials and the number of false trials are obtained for the extracted design examples. The number of true trials and the number of false trials are obtained from the above formulas (3) and (4), as in the second and third embodiments.

The processes after step S68 are the same as the processes after step S28 in FIG. 9 and step S48 in FIG.
As described above, according to the optimum design calculation system according to the present embodiment, a design example similar to the design example in which the device performance value is calculated using the inference analysis based on the memory is taken out. Using the number of true trials and false trials for the extracted design example and the number of true trials and false trials for the design example for which the simulation was performed, the i-th calculated device performance value is up to i−1. The probability of obtaining a value better than the best value among the device performance values calculated in (1) is calculated. By calculating the probability value including the number of true trials and false trials for similar design examples by inference analysis based on memory, not only can the optimal design process be accelerated, but also the accuracy of the probability value can be improved. Therefore, it is possible to calculate the apparatus performance value more appropriately.

<Fifth Embodiment>
FIG. 14 is a block diagram of the optimum design calculation system 10 according to the present embodiment. Since the system configuration is the same as the configuration according to the first embodiment shown in FIG. 1, the description is omitted here, and the optimum design in the optimum design calculation system 10 according to the present embodiment is described with reference to FIG. An overview of the processing will be described.

  The management PC 2 issues an execution start instruction to start the Monte Carlo simulation to the calculation PC 3 according to the information input from the user, and gives the design parameters obtained by the optimization program to the calculation PC 3.

  The calculation PC 3 executes the Monte Carlo simulation program using the given design parameters in accordance with the simulation execution start instruction input from the management PC 2, and from the number of true trials and the number of false trials as in the above embodiment. Calculate device performance values. The number of true trials, the number of false trials, and the device performance value are passed from the calculation PC 3 to the management PC 2.

  The management PC 2 performs analysis using a statistical analysis program, and changes the range of the design parameter from the relationship between the design parameter and the device performance value to limit the range in which a good device performance value can be obtained. Then, the number of trials when performing the Monte Carlo simulation is set for the next design example (design parameter). Then, the simulation result is output to a data file.

  As described above, unlike the first to fourth embodiments, the optimum design calculation system according to the present embodiment performs a Monte Carlo simulation with a predetermined number of trials on a certain design example, and achieves a good apparatus performance value. The design parameter range that can be obtained is calculated. Then, the number of trials corresponding to the calculated design parameter range is calculated, and the next design example is simulated for the number of trials obtained by the calculation. Thereafter, as the same processing is repeated, the setting parameter range is narrowed and the number of trials is increased.

  At the initial stage of the optimization process, the number of trials is kept small, and the calculation time is reduced by calculating the low-accuracy device performance value.In the final stage of the optimization process, instead of increasing the number of trials By limiting the range of design parameters, the calculation speed is increased and a highly accurate apparatus performance value is secured.

FIG. 15 is a flowchart showing an optimum design calculation process according to the fifth embodiment.
First, in step S81, the upper limit value of the number of repeated calculations, the maximum search range of design parameters, the minimum value and the maximum value of the number of trials are input to the optimal design calculation system 10 by the user. Here, as the upper limit value of the number of repeated calculations, the number of design examples to be calculated is set as in the above-described embodiment. The maximum search range of design parameters is, for example, the widest search range among the ranges shown in FIG. The minimum value and the maximum value of the number of trials are respectively the minimum value and the maximum value of the number of times of executing the Monte Carlo simulation for a certain design example.

  In step S82, the management PC 2 sets the maximum search range input by the user as the search range, and in step S83 sets the minimum value input by the user as the number of trials. In step S84, the management PC 2 creates a new data file and initializes a counter for the number of repeated calculations. In step S85, the management PC 2 executes the optimization program to calculate and output design parameter values to be simulated.

In step S86, the calculation PC 3 executes a Monte Carlo simulation using the design parameter value given from the management PC 2, and determines whether the apparatus is normal or abnormal. The number of true trials, the number of false trials, and the device performance value are calculated based on the normal / abnormal device determination result. The simulation, the normal / abnormal device determination, and the device performance value calculation in step S86 are repeated as many times as the number of trials set in the previous step.

  In step S87, the management PC 2 additionally outputs the data received from the calculation PC 3 for the design example for which the Monte Carlo simulation has been performed, that is, the design parameter, the device performance value, the number of true trials, and the number of false trials to the data file. In step S88, the management PC 2 performs regression tree analysis based on the data file, and calculates the optimum region and the change width of the device performance value in the optimum region.

In addition, about the analysis process of step S88, although the regression tree analysis is performed in the Example, it is not restricted to this, It is good also as a structure using other various data mining methods.
In step S89, the management PC 2 changes the search range so as to correspond to the optimum region, and in step S90, the number of trials that can obtain sufficiently good calculation accuracy based on the change width of the device performance value in the optimum region. , Set in the range below the maximum number of trials. In step S91, it is determined whether or not the number of repetition calculations is equal to the upper limit value of the number of repetition calculations. If the number of repetition calculations is less than the upper limit value, the counter of the number of repetition calculations is incremented by 1, and the process proceeds to step S85. Return. Then, referring to the data file, the design parameter estimated to be optimum by the optimization program is set from the design parameter range changed in step S89, and the same processing is executed for the set design example. Go. If the number of repeated calculations is equal to the upper limit value in step S91, the process is terminated.

  As described above, according to the optimal design calculation system according to the present embodiment, a Monte Carlo simulation is performed for a design example to obtain a device performance value, a regression tree analysis or the like is performed to determine the relationship between the design parameter and the device performance value. Ask. Then, the following simulation is performed within a range of design parameters that can provide a good apparatus performance value. Next, when the simulation is performed, the number of trials is increased so as to ensure calculation accuracy while narrowing the range of design parameters. As a result, at the initial stage of optimization, the calculation process can be speeded up by keeping the number of trials small instead of widening the range of design parameters. At the final stage of optimization, the range of design parameters is narrowed, but since a large value is set as the number of trials, it is possible to ensure the accuracy of calculation of device performance values, thereby improving device performance. It becomes possible to improve the reliability of the value.

Regarding the above embodiment, the following additional notes are disclosed.
(Appendix 1)
Computer
A device performance value calculating means for performing a Monte Carlo simulation to obtain the number of true trials and the number of false trials, and calculating a device performance value from the number of true trials and the number of false trials,
The device performance value obtained by performing Monte Carlo simulation on the i-th design example by the device performance calculation means is more than the best device performance value obtained by performing Monte Carlo simulation on the i-1th design example. A probability value calculating means for calculating a probability value to be improved using a binomial distribution based on the number of i-th true trials and false trials;
When the probability value calculated by the probability value calculation means is less than a predetermined threshold, the Monte Carlo simulation by the apparatus performance value calculation means is interrupted, and when the probability value is greater than or equal to a predetermined threshold, the apparatus performance Determination means for continuing the Monte Carlo simulation by the value calculation means,
Optimal design calculation program to function as.
(Appendix 2)
The probability value calculating means takes out a design example in a region to which a design parameter belongs to the i-th design example among the i-th design examples up to the (i−1) th design example by regression tree analysis, and the true trial for the extracted design example The optimal design calculation program according to supplementary note 1, wherein the probability value is calculated based on a value obtained by totaling the number and the number of false trials.
(Appendix 3)
The probability value calculation means takes out a design example in a group to which a design parameter for the i-th design example belongs out of the i-1th design examples by cluster analysis, and the number of true trials for the extracted design example The optimal design calculation program according to appendix 1, wherein the probability value is calculated based on a total value of the number of false trials.
(Appendix 4)
The probability value calculation means extracts a design example similar to the i-th design example from the i-1th design examples by inference analysis based on the memory, and the number of true trials and false trials for the extracted design examples The optimal design calculation program according to appendix 1, wherein the probability value is calculated based on a value obtained by summing the numbers.
(Appendix 5)
The computer,
Design parameter range calculation means for calculating, by data mining, a design parameter range corresponding to a device performance value equal to or less than a predetermined value based on a device performance value obtained by performing a Monte Carlo simulation with a predetermined number of trials on a design example ,
An attempt is made to obtain the change width of the device performance value corresponding to the design parameter range calculated by the design parameter range calculation means, and to set the number of trials for a design example to be subjected to the next Monte Carlo simulation based on the obtained change width Number setting means,
It is determined whether or not a Monte Carlo simulation has been performed for a predetermined number of design examples, and when the Monte Carlo simulation is performed for the predetermined number of design examples, the Monte Carlo simulation by the device performance value calculation unit is performed. A second determination means for ending,
The optimum design calculation program according to appendix 1, wherein the program is further functioned as:
(Appendix 6)
The optimum design according to claim 1, wherein the determination means interrupts the Monte Carlo simulation by the apparatus performance value calculation means when the number of trials of the Monte Carlo simulation for the design example reaches a predetermined upper limit value. Calculation program.
(Appendix 7)
An optimal design calculation method for performing an optimal design calculation by Monte Carlo simulation,
Monte Carlo simulation is performed to obtain the number of true trials and the number of false trials, and the device performance value is calculated from the number of true trials and the number of false trials.
A probability value that the device performance value obtained by performing the Monte Carlo simulation on the i-th design example is better than the best device performance value obtained by performing the Monte Carlo simulation on the i-th design example. , Using a binomial distribution based on the number of i-th true and false trials,
Suspending the Monte Carlo simulation if the calculated probability value is less than a predetermined threshold, and continuing the Monte Carlo simulation if the probability value is greater than or equal to a predetermined threshold;
An optimal design calculation method characterized by including processing.
(Appendix 8)
An optimal design calculation device that performs optimal design calculation by Monte Carlo simulation,
A device performance value calculating means for performing a Monte Carlo simulation to obtain the number of true trials and the number of false trials, and calculating a device performance value from the number of true trials and the number of false trials,
The device performance value obtained by performing Monte Carlo simulation on the i-th design example by the device performance calculation means is more than the best device performance value obtained by performing Monte Carlo simulation on the i-1th design example. A probability value calculating means for calculating a probability value to be improved using a binomial distribution based on the number of i-th true trials and false trials;
When the probability value calculated by the probability value calculation means is less than a predetermined threshold, the Monte Carlo simulation by the apparatus performance value calculation means is interrupted, and when the probability value is greater than or equal to a predetermined threshold, the apparatus performance Determination means for continuing the Monte Carlo simulation by the value calculation means,
An optimum design calculation device characterized by comprising:

It is a block diagram of the optimal design calculation system which concerns on 1st Embodiment. It is a block diagram of the optimal design calculation system which concerns on 1st Embodiment. It is a figure explaining the optimal design by a Monte Carlo simulation. It is a figure explaining a design parameter. It is a figure which shows the example of a data file. It is the flowchart which showed the optimal design calculation process which concerns on 1st Embodiment. It is a figure explaining regression tree analysis. It is a figure which shows the relationship between the design parameter calculated | required by the regression tree analysis, and an apparatus performance value. It is the flowchart which showed the optimal design calculation process which concerns on 2nd Embodiment. It is a figure explaining cluster analysis. It is the flowchart which showed the optimal design calculation process which concerns on 3rd Embodiment. It is a figure explaining inference analysis based on memory. It is the flowchart which showed the optimal design calculation process which concerns on 4th Embodiment. It is a block diagram of the optimal design calculation system which concerns on 5th Embodiment. It is the flowchart which showed the optimal design calculation process which concerns on 5th Embodiment.

Explanation of symbols

1 Optimal design calculation program 2 Management computer (PC for management)
3. Computer for calculation (PC for calculation)

Claims (7)

A computer that performs optimal design calculation by performing Monte Carlo simulation on a design example that is a combination of design parameters ,
Performing Monte Carlo simulations for a plurality of design examples, the operation of each apparatus to obtain a false trial number operation is determined to be abnormal and the determined true number of attempts the system is normal, said vacuum Tries And a device performance value calculating means for calculating a device performance value from the number of false trials,
The best obtained by performing the Monte Carlo simulation on the design example whose device performance value is up to the (i-1) th, obtained by performing Monte Carlo simulation on the i-th design example of the plurality of design examples by the device performance calculating means. the well consists probability value from device performance value, the original binomial distribution and the true number of trials and false number of attempts for best device performance value and the i-th design example of the design example to i-1 th using the A probability value calculation means for calculating
When the probability value calculated by the probability value calculation means is less than a predetermined threshold, the Monte Carlo simulation by the apparatus performance value calculation means is interrupted, and when the probability value is greater than or equal to a predetermined threshold, the apparatus performance Determination means for continuing the Monte Carlo simulation by the value calculation means,
Optimal design calculation program to function as. The probability value calculation means extracts a design example in a region to which a design parameter belongs to the i-th design example from the i-1th design examples by regression tree analysis, and takes out the design example and the i-th design. The optimal design calculation program according to claim 1, wherein the probability value is calculated based on a total value of the number of true trials and the number of false trials for an example. The probability value calculation means takes out a design example in a group to which a design parameter belongs to the i-th design example among the i-1th design examples by cluster analysis, and takes out the design example and the i-th design example. The optimal design calculation program according to claim 1, wherein the probability value is calculated based on a value obtained by totaling the number of true trials and the number of false trials. The probability value calculation means takes out a design example similar to the i-th design example from the i-1th design examples by inference analysis based on memory, and the extracted design examples and the i-th design example The optimal design calculation program according to claim 1, wherein the probability value is calculated based on a total value of the number of true trials and the number of false trials. The computer,
Design parameter range calculation means for calculating, by data mining, a design parameter range corresponding to a device performance value equal to or less than a predetermined value based on a device performance value obtained by performing a Monte Carlo simulation with a predetermined number of trials on a design example ,
An attempt is made to obtain the change width of the device performance value corresponding to the design parameter range calculated by the design parameter range calculation means, and to set the number of trials for a design example to be subjected to the next Monte Carlo simulation based on the obtained change width Number setting means,
It is determined whether or not a Monte Carlo simulation has been performed for a predetermined number of design examples, and when the Monte Carlo simulation is performed for the predetermined number of design examples, the Monte Carlo simulation by the device performance value calculation unit is performed. A second determination means for ending,
The optimal design calculation program according to claim 1, further functioning as: An optimal design calculation method in which a computer performs an optimal design calculation by performing a Monte Carlo simulation on a design example that is a combination of design parameters ,
The computer is
Performing Monte Carlo simulations for a plurality of design examples, the operation of each apparatus to obtain a false trial number operation is determined to be abnormal and the determined true number of attempts the system is normal, said vacuum Tries And the device performance value from the number of false trials,
The device performance value obtained by performing Monte Carlo simulation on the i-th design example of the plurality of design examples is greater than the best device performance value obtained by performing Monte Carlo simulation on the i-1th design example. the well consists probability value, calculated using the original binomial distribution and the true number of trials and false number of attempts for best device performance value and the i-th design example of the design example to (i-1) th,
Suspending the Monte Carlo simulation if the calculated probability value is less than a predetermined threshold, and continuing the Monte Carlo simulation if the probability value is greater than or equal to a predetermined threshold;
Optimal design calculations wherein a call. An optimal design calculation device that performs an optimal design calculation by performing a Monte Carlo simulation on a design example that is a combination of design parameters ,
Performing Monte Carlo simulations for a plurality of design examples, the operation of each apparatus to obtain a false trial number operation is determined to be abnormal and the determined true number of attempts the system is normal, said vacuum Tries And a device performance value calculating means for calculating a device performance value from the number of false trials,
The device performance value obtained by performing the Monte Carlo simulation on the i-th design example of the plurality of design examples by the device performance calculating means is obtained by performing the Monte Carlo simulation on the i-1th design example. the probability value better than the best device performance value, the original binomial distribution and the true number of trials and false number of attempts for best device performance value and the i-th design example of the design example to i-1 th Probability value calculation means to calculate using,
When the probability value calculated by the probability value calculation means is less than a predetermined threshold, the Monte Carlo simulation by the apparatus performance value calculation means is interrupted, and when the probability value is greater than or equal to a predetermined threshold, the apparatus performance Determination means for continuing the Monte Carlo simulation by the value calculation means,
An optimum design calculation device characterized by comprising:

Images