Method for constructing agent model based on convolutional neural network

Abstract

The invention aims to provide a method for constructing an agent model based on a convolutional neural network. The mainstream agent model is generally a model approximated for each sub-target in themulti-target optimization problem, and the agent model is integrally designed for the multi-target optimization problem. In this way, the high efficiency of the agent model in the multi-objective optimization algorithm can be improved. Besides, the convolutional neural network is innovatively used to construct the agent model so that the complex multi-objective optimization problem can be efficiently processed, and the agent model of the problem can still be accurately and efficiently acquired under the condition of large calculated amount. The method comprises the steps of performing Latin cube sampling on a training sample to obtain a training set; distributing a space vector for each sample in the training set; establishing an agent model based on the convolutional neural network by adopting a decomposition thought; and training a multi-objective optimization agent model based on the convolutional neural network.

Description

[0001] 1. Technical field [0002] This method is applied to the multi-objective optimization evolutionary calculation, and uses the convolutional neural network and the surrogate model to design the surrogate model required for the multi-objective optimization problem. This method is aimed at practical engineering problems such as robot design, electronic engineering, etc., and combines the agent model with the multi-objective optimization algorithm to assist the multi-objective optimization evolutionary algorithm to complete the optimization process. [0003] 2. Background technology [0004] 1. Decomposition-based multi-objective optimization problems [0005] The idea of ​​multi-objective decomposition mainly refers to decomposing the multi-objective optimization problem into a certain number of sub-problems, and then optimizing each sub-problem at the same time. In the classic decomposition-based multi-objective optimization evolutionary algorithm (MOEA / D), it uses the so...

AI Technical Summary

Problems solved by technology

The static method has a simple structure and is easy to solve, but its performance is very dependent on the modeling accuracy of the approximate model

With the increase in the complexity of engineering problems and the increase in the size of the design space, it is becoming more and more difficult to achieve the ideal accuracy in one-time modeling in a limited time cost

Images