EE6227-遗传算法概要、优化策略及数学定理

Involved Algorithms

GAs, PSO, Constraint Handling, Evolutionary Strategy, Differential Evolution, Multi-Objective Algorithms and multimodal optimization.

Principle of Evolutionary Algorithms

• Proper encoding is necessary.
• There is a certain index to quantify the fitness of every individual.
• The selection of parents and the reproduction of parents is to some extend random.
• The population of solutions will be better as the number of generations increases.

Introduction to Optimization

The problem can be described as:

$Minimizef(x)overx\in \Omega$

$\Omega isthepermissibleregionofthedecisionvariables,whichisdefinedbyconstraints.$Discrete solution can be considered as an approximate solution to the continuous problem. For example:

$Minimizef(x)=x^{2}where-1\le x\le 1$

The value of x can be discretized for example using 32 bits plus a sign bit, so totally there are $2^{33}$ discrete solutions.

However, it is not efficient enough because the problems are likely to be a large numer of enumerations. Consequently, real-parameter evolutionary algorithms is much better.

Global Optimum v.s. Computational Cost

Due to the computational time, in most cases it is impossible and unnecessary to find the global solution. Algorithms such as the GA or simulated annealing can be used because these approaches make use of randomness to escape from local optimal solutions. However, no method can guarantee to produce the global optimal solution (especially for complex multimodal problems).

Nonlinear Programming Problems

• Can be solved by using methods of calculus, but it is difficult to solve these problems analytically.
• The most popular method is the sequential quadratic programming.
• Iterative numerical algorithm