Today I have lecture class about Differential Evolution or call for short is DE
How DE perform ?
optimality search using mutation, crossover and selection of the best individual along the generation
first step by mutation to generate a mutated vector
then mutated vector and the target vector to create the trial vector
after all selection to keep the better one of the trial vector and target vector
DE Mutation strategies
- x identifies the mutated vector,
- y specifies the number of pair of different vector, and
- z denotes the crossover methods, either binomial or exponential
Binomial Crossover in DE
- if High CR we choose from Vji
- if Low CR we choose from Xji
many formula from here so I skipping on writing about them but you can
look up for it by using this keywords Binomial Crossover , Exponential Crossover
Three Critical Control Parameters of DE
- NP : the population size , suggested 3D - 8D
D is the problem's dimension or the number of decision variables to optimize
[ if NP is increase while F is slightly decreased, the convergence is more likely to occur but generally takes longer. ]
- CR: crossover rate -[0.0,1.0], often high~0.7-0.9
control the influence of parent in generation of offspring
[ more sensitive to a problem's property and complexity, such as the multi-modality ]
- F: Differential scaling factor -(0.0,2.0] often <1.0
Set to 0.5 if unknown.
controls the amplification of differential variation
[ F is more related to the convergence speed ]
Self-Adaptive DE by Qin and Suganthan (NTU) called SaDE
switches between two variants: DE/rand/1/bin and DE/current-to-best/1/bin
Trigonometric Mutation Operation to differential Evolution (2003) by Fan & Lampinen's
Random 3 different vectors, with indices r1, r2, and r3
Opposition-based Differential Evolution by Rahnamayan
- Min and Max is not boundary it's a min and max of current population
- this method is more mutated if do more it not good because it's will produce too much varies
And this is all for today see you again
MySimpleDiary ชีวิตง่ายๆแค่คิดบวก สวัสดีครับ
ไม่มีความคิดเห็น:
แสดงความคิดเห็น