Reduction of fitness evaluations using clustering techniques and neural network ensembles

• US 7,363,281 B2
• Filed: 01/24/2005
• Issued: 04/22/2008
• Est. Priority Date: 01/26/2004
• Status: Expired due to Fees

• Current Assignee: Honda Research Institute Europe GmbH (Honda Motor Company)
• Original Assignee: Honda Research Institute Europe GmbH (Honda Motor Company)
• Inventors: Jin, Yaochu, Sendhoff, Bernhard
• Primary Examiner(s): Vincent; David
• Assistant Examiner(s): Berman; Melissa J
• Application Number:
  US11/042,991
• Publication Number:
  US 20050209982A1
• Time in Patent Office: 1,184 Days
• Field of Search: 706/15
• US Class Current: 706/13
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Abstract

The invention relates to an evolutionary optimization method. First, an initial population of individuals is set up and an original fitness function is applied. Then the offspring individuals having a high evaluated quality value as parents are selected. In a third step, the parents are reproduced to create a plurality of offspring individuals. The quality of the offspring individuals is evaluated selectively using an original fitness function or an approximate fitness function. Finally, the method returns to the selection step until a termination condition is met. The step of evaluating the quality of the offspring individuals includes grouping all offspring individuals in clusters, selecting for each cluster one or a plurality of offspring individuals, resulting in altogether selected offspring individuals, evaluating the selected offspring individuals by the original fitness function, and evaluating the remaining offspring individuals by means of the approximate fitness function.

Claims

1. An evolutionary optimization method for optimizing a hydrodynamic or aerodynamic design, a set of parameters for the hydrodynamic or aerodynamic design represented by individuals, the method comprising the steps of:

(a) setting up an initial population of the individuals representing the set of parameters for the hydrodynamic or aerodynamic design and applying an original fitness function;

(b) selecting offspring individuals having a high evaluated quality value as parents;

(c) reproducing the parents to create a plurality of offspring individuals;

(d) evaluating the quality of the plurality of offspring individuals by selectively using the original fitness function or an approximate fitness function for each offspring individual, the approximation function comprising a neural network ensemble including a plurality of neural networks, evaluating the quality of the plurality of offspring individuals including the steps of;

grouping all of the plurality of offspring individuals into clusters;

selecting for each cluster one or more offspring individuals, resulting in altogether selected offspring individuals;

evaluating the selected offspring individuals by means of the original fitness function to generate original fitness values;

evaluating the remaining offspring individuals by means of the approximate fitness function to generate approximate fitness values;

determining a variance of the approximate fitness values for the remaining offspring individuals; and

replacing the approximate fitness values of the remaining offspring individuals in a cluster with the original fitness value of the selected offspring individual closest to a center of the cluster responsive to the variance of the approximate fitness value being larger than a threshold;

(e) repeating steps (b) through (d) until a termination condition is met; and

(f) storing the offspring individuals representing the set of parameters for the hydrodynamic or aerodynamic design in a computer storage.

2. The method of claim 1, wherein the offspring individuals chosen for evaluation with the original fitness function are the ones closest to the cluster center.

3. The method of claim 1, wherein a k-means method is used to group all of the plurality of offspring individuals into clusters.

4. The method of claim 1, wherein an output of each neural network in the neural network ensemble is weighted and combined to a final output of the neural network ensemble.

5. The method of claim 1, wherein a genetic algorithm with a local search is used to generate the neural network ensemble.

6. The method of claim 1, wherein an evolution strategy is used to optimize weights of the neural network ensemble by minimizing an expected prediction error.

7. The method of claim 1, wherein the method is applied for the optimization of turbine rotor, stator blades, or outlet guide vanes.

8. The method of claim 1, wherein the method is applied for the optimization of aerodynamic properties of vehicles.

9. A computer software program for optimizing a hydrodynamic or aerodynamic design, a set of parameters for the hydrodynamic or aerodynamic design represented by individuals, the computer software program comprising a readable medium structured to store instructions executable by a processor, the instructions, when executed cause the processor to:

(a) set up an initial population of individuals and applying an original fitness function;

(b) select offspring individuals having a high evaluated quality value as parents;

(c) reproduce the parents to create a plurality of offspring individuals;

(d) evaluating the quality of the plurality of offspring individuals by selectively using the original fitness function or an approximate fitness function for each offspring individual, the approximation function comprising a neural network ensemble including a plurality of neural networks, evaluating the quality of the plurality of offspring individuals including the steps of;

grouping all of the plurality of offspring individuals into clusters;

selecting for each cluster one or more offspring individuals, resulting in altogether selected offspring individuals;

evaluating the selected offspring individuals by means of the original fitness function to generate original fitness values;

evaluating the remaining offspring individuals by means of the approximate fitness function to generate approximate fitness values;

determining a variance of the approximate fitness values for the remaining offspring individuals; and

replacing the approximate fitness values of the remaining offspring individuals in a cluster with the original fitness value of the selected offspring individual closest to a center of the cluster responsive to the variance of the approximate fitness value being larger than a threshold;

(e) repeat steps (b) through (d) until a termination condition is met; and

(f) storing the offspring individuals representing the set of parameters for the hydrodynamic or aerodynamic design in a computer storage.