Introduction to Differential Evolution

Rajib Kumar Bhattacharjya

Department of Civil Engineering
Indian Institute of Technology Guwahtai
Differential Evolution
It is a stochastic, population-based optimization algorithm for solving nonlinear
optimization problem

The algorithm was introduced by Storn and Price in 1996

Consider an optimization problem

Minimize ݂ ܺ
Where ܺ = ‫ݔ‬ଵ , ‫ݔ‬ଶ , ‫ݔ‬ଷ , … , ‫ݔ‬஽ , ‫ ܦ‬is the number of variables
 Evolutionary algorithms
Initialize
Mutation Recombination Selection
population

Termination
Next Generation
Criteria

This is a population based algorithm No Yes

Consider a population size of ܰ Optimal
Solution
The population matrix can be shown as
‫ݔ‬௡,௜ = ‫ݔ‬௡,ଵ , ‫ݔ‬௡,ଶ , ‫ݔ‬௡,ଷ , … , ‫ݔ‬௡,஽
௚ ௚ ௚ ௚ ௚

Where, ݃ is the Generation and ݊ = 1,2,3, … ܰ

 Initial population
Initial population is generated randomly between upper lower and upper bound

‫ݔ‬௡,௜ = ‫ݔ‬௡,௜
௅
+ ‫݀݊ܽݎ‬ ∗ ‫ݔ‬௡,௜
௎
− ‫ݔ‬௡,௜
௅
݅ = 1,2,3, … ‫ ܦ‬and ݊ = 1,2,3, … ܰ

Where ‫ݔ‬௜௅ is the lower bound of the variable ‫ݔ‬௜

Where ‫ݔ‬௜௎ is the upper bound of the variable ‫ݔ‬௜

 Mutation
From each parameter vector, select three other vectors ‫ݔ‬௥ଵ௡ , ‫ݔ‬௥ଶ௡ and ‫ݔ‬௥ଷ௡ randomly.
௚ ௚ ௚

Add the weighted difference of two of the vectors to the third

‫ݒ‬௡ = ‫ݔ‬௥ଵ௡ + ‫ݔ ܨ‬௥ଶ௡ − ‫ݔ‬௥ଷ௡

௚ାଵ ௚ ௚ ௚

‫ݒ‬௡
௚ାଵ
is called donor vector

‫ ܨ‬is generally taken between 0 and 1

Recombination

A trial vector ‫ݑ‬௡,௜ is developed from the target vector, ‫ݔ‬௡,௜ , and the donor vector,
௚ାଵ ௚

‫ݒ‬௡,௜
௚ାଵ

‫ݒ‬௡,௜
௚ାଵ
݂݅ ‫ܥ ≤ )(݀݊ܽݎ‬௣ ‫ܫ = ݅ ݎ݋‬௥௔௡ௗ ݅ = 1,2,3, … ‫ ܦ‬and
‫ݑ‬௡,௜ =ቐ ௚
௚ାଵ
‫ݔ‬௡,௜ ݂݅ ‫ܥ > )(݀݊ܽݎ‬௣ ܽ݊݀ ݅ ≠ ‫ܫ‬௥௔௡ௗ ݊ = 1,2,3, … ܰ

‫ܫ‬௥௔௡ௗ is a integer random number between [1,D]

‫ܥ‬௣ is the recombination probability

Selection
The target vector ‫ݔ‬௡,௜ is compared with the trial vector ‫ݑ‬௡,௜ and the one with the
௚ ௚ାଵ

lowest function value is selected for the next generation

‫ݑ‬௡,௜ ݂݅ ݂ ‫ݑ‬௡ < ݂ ‫ݔ‬௡

௚ାଵ ௚ାଵ ௚
‫ݔ‬௡ =ቐ
௚ାଵ
‫ݔ‬௡ ܱ‫݁ݏ݅ݓݎ݄݁ݐ‬
௚

݊ = 1,2,3, … ܰ