Parallel processing design device and parallel processing design method
The parallel processing design device and method address the challenge of high-dimensional design by discarding low-contribution variables and employing Gaussian process regression and Lipschitz continuity to efficiently search for multi-performance executable regions, optimizing vehicle body design.
Patent Information
- Application Number
- JP2020144930
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-08-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2040-08-28
AI Technical Summary
Existing methods face challenges in efficiently searching for a multi-performance executable region due to the exponential increase in search space when dealing with high-dimensional design variables, making it difficult to optimize contradictory performance factors like strength, rigidity, and weight reduction in vehicle body design.
A parallel processing design device and method that discards design variables with low contribution and reduces dimensions using a penalty function and Gaussian process regression, allowing for efficient search of a multi-performance executable region by calculating observation points near the boundary between feasible and infeasible regions.
The method effectively suppresses calculation costs and enhances search efficiency by reducing dimensions and utilizing Lipschitz continuity-based penalty functions, enabling high-dimensional design optimization and accurate derivation of multi-performance executable regions.
Smart Images

Figure 0007584256000011 
Figure 0007584256000012 
Figure 0007584256000013
Abstract
Description
[Technical field]
[0001] The present invention relates to a parallel processing design device and a parallel processing design method for designing a structure such as a vehicle body. [Background technology]
[0002] In the design of structures, one of the important issues is to establish a design methodology that can optimize multiple performance factors (multiple performance factors) that may sometimes be contradictory, such as strength, rigidity, weight reduction, and vibration suppression. Methods for optimizing multiple performance factors simultaneously in parallel using computer simulations are being researched.
[0003] In Non-Patent Document 1, when designing a product, a region in which target performance is established is found under multiple known constraint conditions. As a solution search method, Bayesian optimization is used to search for design variables that maximize the Entropy Search (ES), which is an acquisition function. Then, a method is disclosed in which ES is used to search only limited observation points and model the feasible region with sufficient accuracy. [Prior art documents] [Patent documents]
[0004] [Non-Patent Document 1] Philipp Hennig and Christian J. Schuler, “Entropy Search for Information-Efficient Global Optimization,” Journal of Machine Learning Research, vol. 13, no. Jun, pp. 1809-1837, 2012. Summary of the Invention [Problem to be solved by the invention]
[0005] However, the invention described in Non-Patent Document 1 has a problem in that when there are many design variables, such as 30 or more dimensions, the search space increases exponentially, making efficient search difficult.
[0006] In consideration of the above, an object of the present invention is to provide a parallel processing design device and a parallel processing design method capable of efficiently searching for a multi-performance feasible region. [Means for solving the problem]
[0007] The parallel processing design device according to claim 1 includes an observation point calculation unit that, when the order of a plurality of design variables exceeds a predetermined order, discards design variables having little contribution to design, and calculates, for each of the plurality of design variables, a plurality of observation points for searching a region in which performance related to each of the plurality of design variables that has become equal to or less than the predetermined order is feasible, using an acquisition function related to the search of the feasible region and a penalty function that expresses a region in which the observation point cannot be taken in the acquisition function; a probability distribution calculation unit that calculates, for each of the plurality of performances, a probability distribution in which each of the plurality of performances is feasible at the calculated plurality of observation points; and a multi-performance feasible region output unit that outputs a sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible region.
[0008] The parallel processing design device according to claim 1 suppresses the exponential increase in calculation cost by discarding dimensions that have a small contribution to calculation and reducing the dimensions, thereby enabling substantial support for high-dimensional designs.
[0009] Moreover, the parallel processing design device according to claim 1 makes it possible to improve the efficiency of searching for observation points by using a penalty function that expresses an area in which an observation point cannot be taken in the acquisition function.
[0010] The parallel processing design device according to claim 2 is the parallel processing design device according to claim 1, wherein the observation point calculation unit rejects design variables for which the variance of multiple data items existing in a design space indicated by each of the multiple design variables is equal to or less than a predetermined threshold value.
[0011] The parallel processing design device according to claim 2 reduces the dimensions of the design variables by discarding design variables whose variance of a plurality of data is equal to or less than a predetermined threshold value.
[0012] A parallel processing design device according to a third aspect of the present invention is the parallel processing design device according to the first or second aspect, wherein the acquisition function is used for searching the vicinity of the boundary between the feasible region and the infeasible region.
[0013] The parallel processing design device according to claim 3 can acquire an observation point that exists near the boundary between a feasible region and an infeasible region.
[0014] A parallel processing design device according to claim 4 is a parallel processing design device according to any one of claims 1 to 3, wherein the penalty function is defined based on Lipschitz continuity of a black-box function modeled using Gaussian process regression.
[0015] The parallel processing design device according to claim 4 can limit the area from which observation points are obtained based on Lipschitz continuity, which indicates that there is an upper limit to the absolute value of the gradient of a line connecting any two points on a graph.
[0016] The parallel processing design device according to claim 5 is a parallel processing design device according to any one of claims 1 to 4, wherein the observation point calculation unit determines the point that maximizes the acquisition function as the first observation point, and determines the point that maximizes the product of the acquisition function and the penalty function as the other observation point.
[0017] The parallel processing design device according to claim 5 can simultaneously obtain a plurality of observation points by calculating the maximum value of acquisition functions that can be simultaneously calculated in parallel processing and the maximum value of the product of an acquisition function and a penalty function.
[0018] The parallel processing design method according to claim 6 includes an observation point calculation step of, when the order of the plurality of design variables exceeds a predetermined order, discarding design variables having little contribution to design, and calculating, for each of the plurality of design variables whose order is equal to or less than the predetermined order, a plurality of observation points for searching a feasible region for performance related to each of the plurality of design variables, using an acquisition function related to the search of the feasible region and a penalty function expressing a region in which the observation point cannot be taken in the acquisition function; a probability distribution calculation step of calculating, for each of the plurality of performances, a probability distribution in which each of the plurality of performances is feasible at the calculated plurality of observation points; and a multi-performance feasible region output step of outputting a sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible region.
[0019] The parallel processing design method according to claim 6 suppresses the exponential increase in calculation cost by discarding dimensions that have a small contribution to calculation and reducing the dimensions, thereby enabling substantial support for high-dimensional designs.
[0020] In addition, the parallel processing design method according to claim 6 makes it possible to improve the efficiency of searching for an observation point by using a penalty function that expresses an area in which an observation point cannot be taken in the acquisition function.
[0021] A parallel processing design method according to claim 7 is the parallel processing design method according to claim 6, in which the observation point calculation step discards design variables for which the variance of multiple data items existing in a design space indicated by each of the multiple design variables is equal to or less than a predetermined threshold value.
[0022] The parallel processing design method according to claim 7 reduces the dimensions of the design variables by discarding design variables whose variance of a plurality of data is equal to or less than a predetermined threshold value.
[0023] A parallel processing design method according to an eighth aspect of the present invention is the parallel processing design method according to the sixth or seventh aspect, wherein the acquisition function is used for searching the vicinity of the boundary between the feasible region and the infeasible region.
[0024] The parallel processing design method according to claim 8 can acquire an observation point that exists near the boundary between a feasible region and an infeasible region.
[0025] The parallel processing design method according to claim 9 is the parallel processing design method according to any one of claims 6 to 8, wherein the penalty function is defined based on Lipschitz continuity of a black-box function modeled using Gaussian process regression.
[0026] The parallel processing design method according to claim 9 can limit the area for obtaining observation points based on Lipschitz continuity, which means that there is an upper limit to the absolute value of the gradient of a line connecting any two points on a graph.
[0027] The parallel processing design method according to claim 10 is a parallel processing design method according to any one of claims 6 to 9, wherein the observation point calculation step determines a point that maximizes the acquisition function as a first observation point, and determines a point that maximizes the product of the acquisition function and the penalty function as another observation point.
[0028] The parallel processing design method according to claim 10 can simultaneously obtain a plurality of observation points by calculating the maximum value of the acquisition function that can be simultaneously calculated in the parallel processing and the maximum value of the product of the acquisition function and the penalty function. Effect of the Invention
[0029] As described above, the parallel processing design apparatus and parallel processing design method according to the present invention make it possible to efficiently search for a multi-performance feasible region. [Brief description of the drawings]
[0030] [Figure 1] 1 is a block diagram showing an example of a specific configuration of a parallel processing design apparatus according to an embodiment of the present invention; [Diagram 2] FIG. 2 is a functional block diagram of a CPU of a parallel processing design apparatus according to an embodiment of the present invention. [Diagram 3] FIG. 1 is a conceptual diagram of a system design according to an embodiment of the present invention. [Figure 4] 1 is an example of a flowchart relating to deriving a feasible region of active learning according to an embodiment of the present invention. [Diagram 5] 5 is a flowchart showing details of step 406 in FIG. 4. [Figure 6] FIG. 1 is a schematic diagram illustrating an example of regression of a one-dimensional function using Gaussian process regression. [Figure 7] 1 is an example of a probability distribution of constraints derived by Gaussian process regression. [Figure 8] FIG. 2 is a conceptual diagram of dimensionality reduction according to an embodiment of the present invention. [Figure 9] FIG. 13 is a schematic diagram showing an example of a calculation result using an acquisition function α(x). [Figure 10] FIG. 1 is an explanatory diagram showing an example of Lipschitz continuity. [Figure 11] (A) is an explanatory diagram showing a case where the point at which the acquisition function is maximized is obtained as the first (initial) observation point, (B) is a case where the second (next) observation point is obtained using a penalty function, and (C) is a diagram showing a case where the third (further subsequent) observation point is obtained. [Figure 12] FIG. 11 is a schematic diagram comparing the F1 score of the conventional method with the F1 score of the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0031] A parallel processing design device and a parallel processing design method according to this embodiment will be described below with reference to Fig. 1. Fig. 1 is a block diagram showing an example of a specific configuration of a parallel processing design device 10 according to an embodiment of the present invention.
[0032] The parallel processing design device 10 is configured to include a computer 30. The computer 30 includes a CPU 32, a ROM 34, a RAM 36, and an input / output port 38. As an example, the computer 30 is preferably a model capable of executing advanced arithmetic processing at high speed, such as an engineering workstation or a supercomputer.
[0033] In the computer 30, the CPU 32, ROM 34, RAM 36, and input / output port 38 are connected to one another via various buses such as an address bus, a data bus, and a control bus. The input / output port 38 is connected to various input / output devices, such as a display 40, a mouse 42, a keyboard 44, a hard disk (HDD) 46, and a disk drive 50 that reads information from various disks (e.g., CD-ROM, DVD, etc.) 48.
[0034] Further, a network 52 is connected to the input / output port 38, and information can be exchanged with various devices connected to the network 52. In this embodiment, a data server 56 to which a database (DB) 54 is connected is connected to the network 52, and information can be exchanged with the DB 54.
[0035] Data related to parallel processing design and the like are stored in advance in the DB 54. Information may be stored in the DB 54 by the computer 30 or the data server 56, or may be registered by another device connected to the network 52.
[0036] In this embodiment, it is described that parallel processing design data and the like are stored in DB 54 connected to data server 56, but the information in DB 54 may also be stored in an external storage device such as HDD 46 built into computer 30 or an external hard disk.
[0037] A parallel processing design program for parallel processing design is installed in the HDD 46 of the computer 30. In this embodiment, the CPU 32 executes the parallel processing design program to execute the parallel processing design. The CPU 32 also displays the processing results of the parallel processing design program on the display 40. There are several methods for installing the parallel processing design program of this embodiment in the computer 30. For example, the parallel processing design program is stored on a CD-ROM or DVD together with a setup program, and the disk is inserted in the disk drive 50, and the setup program is executed by the CPU 32 to install the parallel processing design program in the HDD 46. Alternatively, the parallel processing design program may be installed in the HDD 46 by communicating with another information processing device connected to the computer 30 via a public telephone line or a network 52.
[0038] FIG. 2 shows a functional block diagram of the CPU 32 of the parallel processing design device 10. Various functions realized by the CPU 32 of the parallel processing design device 10 executing the parallel processing design program will be described. The parallel processing design program includes a simulation function for acquiring an observed value for each of a plurality of performances by simulation, an observed value complementing function for complementing the acquired discrete observed value and outputting a continuous predicted value and a prediction error for each of the plurality of performances, an observation point calculation function for calculating a plurality of observation points for searching an area in which each of the plurality of performances is feasible, a probability distribution calculation function for calculating a probability distribution in which each of the plurality of performances is feasible at the plurality of observation points, and a multi-performance feasible area output function for outputting the sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible area. By the CPU 32 executing the parallel processing design program having each of these functions, the CPU 32 functions as a simulation unit 72, an observed value complementing unit 74, an observation point calculation unit 76, a probability distribution calculation unit 78, and a multi-performance feasible area output unit 80, as shown in FIG. 2.
[0039] In parallel processing design, a feasible region is derived in which multiple different, and sometimes mutually contradictory, performances of a design object, such as strength, rigidity, weight reduction, and vibration suppression, can be optimized. The feasible region is generally derived in the following procedure. First, a model between performance-related variables and the response of the variables is defined. Next, feasible region candidates are obtained by random sampling on the previously defined model. Then, samples that satisfy the response constraints are extracted from the obtained results. Although this procedure is simple in principle, there is a problem in that the space to be searched increases exponentially as the dimensions of the design variables increase, resulting in enormous computational costs.
[0040] In this embodiment, the exponential increase in calculation cost can be suppressed by using dimensionality reduction that discards dimensions that have a small contribution to calculation.
[0041] In addition, in this embodiment, by expressing the feasible region as a probability distribution, it is possible to independently obtain the feasible region of each performance in the multi-performance, and by multiplying the feasible region of each performance, it is possible to easily derive the multi-performance feasible region. Even when a new constraint is imposed, it is possible to independently derive a probability distribution related to the new constraint, and multiply the derived probability distribution by the multi-performance feasible region described above, thereby deriving the multi-performance feasible region that takes the new constraint into consideration.
[0042] 3 is a conceptual diagram of a system design in this embodiment. In FIG. 3(1), a feasible region that satisfies the constraints of performance 1, performance 2, and performance 3 is defined as a probability distribution Pr(C k (x)) (k=1, ..., K). k Since (x)) is the probability of realizing each performance, it is as follows. C k (x) is a Boolean function with variable x. 0≦Pr(C k (x)≦1
[0043] Figure 3 (2) shows an example of a probabilistic representation of the multi-performance feasible region. As mentioned above, the probability distribution of each of performances 1 to 3 is Pr(C k Since (x)), the multi-performance feasible region, which is the region where performances 1 to 3 are simultaneously realized, is expressed by the sum of the probability distributions related to each of the performances. In actual design, the design variables may have 30 or more dimensions, so the search space related to deriving the probability distribution increases exponentially, making efficient search difficult. In this embodiment, as described later, the calculation load is reduced by dimensionality reduction, which discards design variables that have a low contribution in the design.
[0044] FIG. 4 is an example of a flowchart related to derivation of a feasible region of active learning according to this embodiment. The CPU 32 reads out a program from the ROM 34 or the disk drive 50, expands it in the RAM 13, and executes it, thereby performing the processing shown in FIG. 4. The processing shown in FIG. 4 requires, as a prerequisite, inputting conditions for realizing multi-performance. As an example of the conditions to be input, the constraint function g for each performance in multi-performance as follows: k (x). The subscript k in the constraint function is the degree of the constraint condition and is also an index for each performance in the multi-performance, and K is the maximum degree of the variables that make up the multi-performance. JPEG0007584256000001.jpg868
[0045] The probability that each performance in a multi-performance system is feasible is given by the following formula (1). k As mentioned above, (x) is a Boolean function with x as a variable. k is a small positive value indicating the tolerance. JPEG0007584256000002.jpg10129
[0046] In addition, the multi-performance feasible region, which is the region in which performance k (k=1, 2, ..., K) is simultaneously realized, is expressed as the sum of the probabilities related to each performance, as shown in the following equation (2). JPEG0007584256000003.jpg2291
[0047] In step 400, an experimental plan is generated to derive a region in which multi-performance y, such as strength, rigidity, weight reduction, and vibration suppression, can be achieved for a variable x, such as the position of the structure or the moment of inertia acting on the structure. Also in step 400, the design defined by the experimental plan is evaluated by a simulation such as CAE (Computer Aided Engineering). By the simulation such as CAE, a value of, for example, y corresponding to the variable x is discretely derived as an observation value.
[0048] In step 402, a prediction model is constructed. In this embodiment, a method called Gaussian process regression is used, which can interpolate and predict the observed value y by considering the correlation of the observed value y with respect to the variable x. Gaussian process regression generally determines the correlation between the variable x and the observed value y using a Gaussian distribution, and is characterized by being able to not only stochastically continuously interpolate the discrete observed value y, but also to calculate the prediction error.
[0049] FIG. 6 is a schematic diagram showing an example of regression of a one-dimensional function using Gaussian process regression. FIG. 6 shows a curve 102 assuming that the observed values 100A, 100B, 100C, 100D, 100E, and 100F are continuous. As shown in FIG. 6, the curve 102 is a continuous function corresponding to the variable x. Since it is a continuous function, not only can the discrete data be interpolated and predicted, but also differentiation with respect to the variable x is possible. In FIG. 6, a region of a prediction error 106 exists around the curve 102. The region of the prediction error 106 becomes narrow when the reliability of the predicted value indicated by the curve 102 is high, but becomes wide when the reliability is low. FIG. 6 shows an inequality constraint value 104 of y=0 as an example. In this embodiment, a region in which the response y is smaller than the inequality constraint value 104 is defined as a feasible region.
[0050] In this embodiment, the cumulative density distribution (CDF) for the variable x is calculated using the interpolation of the discrete data obtained by Gaussian process regression, the prediction error 106, and the inequality constraint value 104.
[0051] FIG. 7 is an example of a probability distribution of a constraint condition derived by Gaussian process regression. In FIG. 7, the horizontal axis is the variable x, and the vertical axis is the value of the cumulative density distribution. The cumulative density distribution 110 in FIG. 7 indicates the probability that the value of y in FIG. 6 is equal to or less than the inequality constraint value 104. When the cumulative density distribution Φ is used, the above formula (1) becomes the following formula (1A), and the probability distribution Pr(C k In the following formula, b is a numerical value related to the lower boundary 134 described later. In addition, in the following formula, σ(x) is a prediction deviation calculated in the calculation process of Gaussian process regression, and μ(x) is a prediction average value. JPEG0007584256000004.jpg18157
[0052] In this embodiment, the results of Gaussian process regression (predicted value, prediction error) are used to calculate observation points for searching the multi-performance feasible region in the design space, but if the order of the design variables is high, the search space will increase exponentially. In step 404, it is determined whether the order of the design variables of the prediction model constructed using Gaussian process regression is equal to or less than a predetermined dimension. The predetermined dimension is, for example, any one of 10 to 30 dimensions.
[0053] If it is determined in step 404 that the degree of the design variables of the prediction model is equal to or less than the predetermined dimension, the procedure proceeds to step 406. If it is determined in step 404 that the degree of the design variables of the prediction model is not equal to or less than the predetermined dimension, the procedure proceeds to step 408.
[0054] In step 406, an observation point is searched for using an acquisition function (Acquisition Function) (to be described later) and a penalty function (to be described later), and the searched new observation point is added to the machine learning teacher data to update the teacher data. Details of the process in step 406 will be described later with reference to FIG. 5 etc.
[0055] In step 408, low-contribution dimensions are discarded. In this embodiment, VAR (Variance Average Relevance) is used to detect dimensions with low design contributions. As shown in FIG. 8, VAR is a method for detecting a variable x in an i-dimensional (i=1, 2, ..., K) design space. j The variance of the posterior mean in the J direction is calculated for each of the variables, and each calculated variance is compared with a predetermined threshold. If the variance value is below the threshold, it is deemed to be insignificant, and the i-dimensional variable x j The dimension reduction is performed by discarding the above. The predetermined threshold is determined taking into consideration the computational load. For example, if there is no problem even if the computational load is increased slightly, the threshold is set low, and if it is desired to suppress the computational load, the threshold is set high.
[0056] In step 410, a dimensionally reduced prediction model is constructed. More specifically, the above-mentioned Gaussian process regression is used to calculate the complementation and prediction error of the observed value y for the dimensionally reduced variable x.
[0057] In step 412, it is determined whether the termination conditions for the process have been met. The termination conditions in step 412 are defined by the following equations (3), (4), and (5), respectively, and the convergence of the calculation is determined using the termination condition shown in equation (5). Equation (5) represents the ratio of the area where the determination of validity / falseness has not been adequately made to the entire area, that is, the ratio of the area related to the design where the observation points have not been calculated. δ in equation (4) k is a small positive value indicating the allowable error. Also, ε on the right side of equation (5) is a threshold indicating the termination condition, and is a small positive value. If the termination condition of the process is satisfied in step 412, the procedure proceeds to step 416. If the termination condition of the process is not satisfied in step 412, the procedure proceeds to step 414. JPEG0007584256000005.jpg18121 JPEG0007584256000006.jpg33129
[0058] In step 414, the constructed dimensionality reduction model is output and is subjected to machine learning in step 406.
[0059] In step 416, a model showing the feasible region is output and the process ends. The feasible region of each performance is expressed as a probability distribution Pr(C k (x)), the multi-performance feasible region that satisfies all the performance constraints can be easily obtained as a joint probability distribution, as shown in the above equation (2). JPEG0007584256000007.jpg2291
[0060] Figure 5 is a flowchart detailing step 406 of Figure 4. Step 500 performs batch machine learning to improve the machine learning process by simultaneously and in parallel sampling multiple observations and updating a predictive model based on Gaussian process regression.
[0061] In the conventional prediction model, the acquisition function and the regression model are updated every time an observation point is obtained. However, in this embodiment, the first observation point is obtained by maximizing the acquisition function, and the remaining (other) observation points are obtained by defining a penalty function based on Lipschitz continuity and obtaining a point where the simultaneous probability of the acquisition function and the penalty function is maximized. From the above, it is possible to simultaneously obtain multiple points in parallel processing.
[0062] The acquisition function is α(x) as ES (Entropy Search) defined based on Gaussian process regression. α(x) is used to search for observation points near the boundaries (upper layer boundary 124, lower layer boundary 134) between the feasible region 120 and the infeasible regions 122, 132. FIG. 9 is a schematic diagram showing an example of a calculation result using α(x). In FIG. 9, a feasible observation point 126 and an infeasible observation point 128 exist near the upper layer boundary 124 and the lower layer boundary 134, so α(x) is suitable for searching near the upper layer boundary 124 or the lower layer boundary 134. α(x) is expressed by the following formula (4). H(p(f(x)) in formula (6) is entropy (Shannon information). JPEG0007584256000008.jpg56154
[0063] The penalty function is formulated using Lipschitz continuity. A Lipschitz continuous function that has Lipschitz continuity has a limited speed of change, and there exists a finite real number that is appropriate for the speed of change. Therefore, it is considered that there is an upper limit to the absolute value of the slope of the line connecting any two points on the graph of the function, and this upper limit is defined as the Lipschitz constant L of the function.
[0064] Fig. 10 is an explanatory diagram showing an example of Lipschitz continuity. Fig. 10 shows a curve 140 of an evaluation function, and evaluation points 142A, 142B, and 142C exist on the curve 140. Then, feasible region boundaries 144 and 146 are defined, and straight lines 152A+ and 152A- with a slope of ±L that pass through evaluation point 142A, straight lines 152B+ and 152B- with a slope of ±L that pass through evaluation point 142B, and straight lines 152C+ and 152C- with a slope of ±L that pass through evaluation point 142C are defined.
[0065] As described above, in Lipschitz continuity, when a straight line with a slope of ±L is drawn between any two points on the function, the absolute value of the slope at any two points on the function is smaller than L. Therefore, there is no possibility that the evaluation function can be a feasible region in the region 160A defined by the intersection of the lines 152A+, 152A- with the feasible region boundary 146, the region 160B defined by the intersection of the lines 152B+, 152B- with the feasible region boundary 146, and the region 160C defined by the intersection of the lines 152C+, 152C- with the feasible region boundary 146. Conversely, there is a possibility that a feasible region exists in the regions 162A, 162B, 162C, and 162D excluding the regions 160A, 160B, and 160C.
[0066] In this embodiment, the penalty function is defined based on Lipschitz continuity to express the regions 160A, 160B, and 160C where observation points cannot be taken in the same acquisition function. The penalty function is defined as the following formula (7) using the Lipschitz continuity of the evaluation function (black box function) modeled using the above-mentioned Gaussian process regression, and the mean μ and deviation σ calculated from the posterior distribution. α and β in the following formula (7) are values that indicate the boundaries of the feasible region. JPEG0007584256000009.jpg51164
[0067] In this embodiment, as described above, the first observation point that maximizes the acquisition function is obtained, and the remaining observation points are obtained by obtaining points that maximize the simultaneous probability of the acquisition function and the penalty function.
[0068] Figure 11 is an explanatory diagram showing an example of calculating multiple observation points, where Figure 11(A) shows the case where the point at which the acquisition function is maximum is obtained as the first (initial) observation point, Figure 11(B) shows the case where the second (next) observation point is obtained using the penalty function shown in equation (7), and Figure 11(C) shows the case where the third (further next) observation point is obtained.
[0069] In FIG. 11A, the point at which a curve 170 of the acquisition function α(x) defined by equation (6) is maximum is acquired as the first observation point 172.
[0070] In FIG. 11(B), a second observation point 178 is obtained using the penalty function φ1(x) (j=1 in equation (7)) defined in equation (7). Specifically, a second observation point 178 is extracted at which a curve 176 indicated by the product of the acquisition function α(x) indicated by curve 170 and the penalty function φ1(x) indicated by curve 174 is maximized. By multiplying the acquisition function α(x) by the penalty function φ1(x), it is possible to obtain observation points from areas 162A, 162B, 162C, and 162D excluding areas 160A, 160B, and 160C in FIG. 10 where no observation points can be obtained.
[0071] In FIG. 11(C), the penalty function is updated to penalty function φ2(x) (j=2 in equation (7)) shown by curve 180, and a third observation point 184 is extracted where curve 182 shown by the product obtained by multiplying the product of the acquisition function α(x) and the penalty function φ1(x) by the updated penalty function φ2(x) is maximized.
[0072] In this embodiment, the acquisition function α(x) is a penalty function φ jBy successively multiplying (x) (j=1, 2, 3, ...) and extracting the maximum product, multiple observation points can be searched.
[0073] In step 502, the acquired information of the observation point is mapped to the prediction model before the dimensionality reduction, thereby updating the teacher data for machine learning, and the process returns.
[0074] Equation (8) is a 30-dimensional Morris function, and the problem is to search for a feasible region where the function range is in the range of 10% to 20%. And equation (9) is a calculation formula for the F1 score based on precision and recall. JPEG0007584256000010.jpg69164
[0075] 12 is a schematic diagram showing the F1 score 190 of the feasible region search by the conventional method without dimensionality reduction calculated by equations (8) and (9), and the F1 score 192 of the feasible region search by the parallel processing design method according to this embodiment. The F1 score is 1.0 when the predicted region and the correct answer region match. As shown in FIG. 12, at the time of 60 calculations, the F1 score 190 of the conventional method is about 0.8, but the F1 score 192 of this embodiment is 1.0, and it can be said that the dimensionality reduction according to this embodiment is effective.
[0076] As described above, in this embodiment, when a multi-performance feasible region is obtained using machine learning, dimensions that have a small contribution to the calculations are discarded to reduce the dimensions, thereby suppressing the exponential increase in calculation cost and making it possible to practically handle high-dimensional designs.
[0077] In addition, in this embodiment, a penalty function defined using Lipschitz continuity is added to the acquisition function based on Gaussian process regression, thereby enabling parallel processing of the observation point search and further improving the efficiency of the search process. In this embodiment, the first observation point is obtained as the point at which the acquisition function is maximized, and the other observation points are obtained using the penalty function. The maximum value of the acquisition function and the maximum value of the product of the acquisition function and the penalty function can be calculated simultaneously in parallel processing, so multiple observation points can be obtained simultaneously.
[0078] This embodiment is capable of practically dealing with high-dimensional designs, and can efficiently calculate solutions even when there are many design variables, such as design specifications or control constants of components that are directly linked to the performance for which a viable range is desired in vehicle development.
[0079] In addition, this embodiment obtains multiple observation points through parallel processing, making it possible to analyze the feasible region in a shorter time than in the past. In cases where the feasible region analysis must be carried out within a limited number of test man-hours, the processing time can be reduced by processing the data simultaneously in multiple processes rather than in a single process.
[0080] According to this embodiment, for example, it is possible to derive engine mount specifications that satisfy the multi-performance feasible region. As an example, in a specification study of engine mount characteristics, engine mount shape, and the like, as many as 20 variables (dimensions) must be considered for one engine mount. However, according to this embodiment, by reducing the design dimensions and performing parallel processing of observation point acquisition, it is possible to quickly and accurately derive the multi-performance feasible region. As a result, it becomes possible to efficiently search for the multi-performance feasible region.
[0081] In this embodiment, the acquisition function used is ES, which is suitable for searching for observation points near the boundaries between the feasible region 120 and the infeasible regions 122, 132, but it is also possible to use PoF (Probability of Feasibility), which is suitable for searching for observation points within the feasible region 120.
[0082] The "acquisition function" described in the claims corresponds to the "acquisition function α(x)" described in the detailed description of the invention in the specification.
[0083] In addition, the processing executed by the CPU after reading the software (program) in each of the above embodiments may be executed by various processors other than the CPU. In this case, examples of the processor include a PLD (Programmable Logic Device) such as an FPGA (Field-Programmable Gate Array) whose circuit configuration can be changed after manufacture, and a dedicated electric circuit such as an ASIC (Application Specific Integrated Circuit) which is a processor having a circuit configuration designed exclusively for executing a specific processing. In addition, the processing may be executed by one of these various processors, or may be executed by a combination of two or more processors of the same or different types (for example, a plurality of FPGAs, a combination of a CPU and an FPGA, etc.). In addition, the hardware structure of these various processors is, more specifically, an electric circuit that combines circuit elements such as semiconductor elements.
[0084] In addition, in each of the above embodiments, the program is described as being pre-stored (installed) in the disk drive 50 or the like, but the present invention is not limited to this. The program may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network.
[0085] (Additional note 1) Memory, at least one processor coupled to the memory; Including, The processor, When the degree of a plurality of design variables exceeds a predetermined degree, a design variable having a small contribution to design is discarded, and a plurality of observation points for searching a region in which performance related to each of the plurality of design variables that has become equal to or less than the predetermined degree is feasible are calculated for each of the plurality of design variables using an acquisition function related to the search for the feasible region and a penalty function that expresses a region in which the observation points cannot be taken in the acquisition function; calculating a probability distribution for each of the plurality of performances that each of the plurality of performances is feasible at the calculated plurality of observation points; outputting a sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible region; It is configured as follows: Parallel Processing Design Device. [Explanation of symbols]
[0086] 10 Parallel processing design device 30 Computers 32 CPU 34 ROM 36 RAM 38 I / O ports 40 Display 42 Mouse 44 Keyboard 50 disk drives 52 Network 56 Data Server 74 Observation Completion Section 76 Observation point calculation section 78 Probability distribution calculation unit 80 Multi-performance feasible region output unit 100A, 100B, 100C, 100D, 100E, 100F Observations 102 curve 104 Inequality Constraint Values 106 Prediction Error 110 Cumulative density distribution 120 Feasible Region 122 Infeasible Region 124 Upper Boundary 126 viable observation points 128, 132 Infeasible area 134 Lower boundary 170 curve 172 observation points 174, 176 curve 178 observation points 180, 182 curve 184 observation points
Claims
1. an observation point calculation unit that, when the order of a plurality of design variables exceeds a predetermined order, discards design variables having a small contribution to design, and calculates, for each of the plurality of design variables that have become equal to or less than the predetermined order, a plurality of observation points for searching a region in which performance related to each of the plurality of design variables is feasible, using an acquisition function related to the search for the feasible region and a penalty function that expresses a region in which the observation points cannot be taken in the acquisition function; a probability distribution calculation unit that calculates, for each of the plurality of performances, a probability distribution that each of the plurality of performances is feasible at the calculated plurality of observation points; a multi-performance feasible region output unit that outputs a sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible region; A parallel processing design device comprising:
2. 2. The parallel processing design device according to claim 1, wherein the observation point calculation unit rejects design variables in which the variance of a plurality of pieces of data present in the design space indicated by each of the plurality of design variables is equal to or less than a predetermined threshold value.
3. 3. The parallel processing design device according to claim 1, wherein the acquisition function is used for searching the vicinity of the boundary between the feasible region and the infeasible region.
4. 4. The parallel processing design apparatus according to claim 1, wherein the penalty function is defined based on Lipschitz continuity of a black-box function modeled using Gaussian process regression.
5. 5. A parallel processing design device according to claim 1, wherein the observation point calculation unit determines a point that maximizes the acquisition function as a first observation point, and determines a point that maximizes the product of the acquisition function and the penalty function as another observation point.
6. an observation point calculation step of, when the order of a plurality of design variables exceeds a predetermined order, discarding design variables having a small contribution to design, and calculating a plurality of observation points for searching a region in which performance related to each of the plurality of design variables that has become equal to or less than the predetermined order is feasible, for each of the plurality of design variables, using an acquisition function related to the search for the feasible region and a penalty function that expresses a region in which the observation points cannot be taken in the acquisition function; a probability distribution calculation step of calculating, for each of the plurality of performances, a probability distribution that each of the plurality of performances is feasible at the plurality of calculated observation points; a multi-performance feasible region output step of outputting a sum of the probability distributions calculated for each of the plurality of performances as a multi-performance feasible region; A parallel processing design method including:
7. 7. The parallel processing design method according to claim 6, wherein the observation point calculation step discards design variables whose variances of a plurality of data items existing in the design space represented by each of the plurality of design variables are equal to or less than a predetermined threshold value.
8. 8. The parallel processing design method according to claim 6, wherein the acquisition function is used for searching the vicinity of the boundary between the feasible region and the infeasible region.
9. The method for designing parallel processing according to any one of claims 6 to 8, wherein the penalty function is defined based on Lipschitz continuity of a black-box function modeled using Gaussian process regression.
10. 10. The parallel processing design method according to claim 6, wherein the observation point calculation step sets a point that maximizes the acquisition function as a first observation point, and sets a point that maximizes the product of the acquisition function and the penalty function as another observation point.
Citation Information
Patent Citations
System for creating parameter information, system for estimating yields, program and recording medium
JP2010160787A
Optimization device and method
JP2015225635A
Input parameter search device, input parameter search method, and input parameter search program
JP2018073360A
Optimization device, simulation system and optimization method
JP2020027370A
