EVOLVE

Version
Jan 17, 1993

Description
EVOLVE is a genetic search algorithm that can be used to optimize problems with a mix of continuous, integer and discrete design variables. Constraints are considered by using an exterior penalty function. It has a capability to improve global search (called sharing) by penalizing designs that are near to other good designs.

References
EVOLVE Manual

Control Parameters

Name Default Value Description
Crossover
CrossoverRate 0.8 Probability of crossover. This is typically between 0.6 and 0.9 and valid range is [0, 1].
DirectedCrossover 0 Generation at which directed crossover is to be started. 0 disables directed crossover. Option accepts positive integer values.
DirectedCrossoverGenerations 1 Number of previous generations over which information of positional gain will be used to direct the crossover. Use only if DirectedCrossover > 1. Option must have an integer value greater than 0.
Optimization Parameters
Accuracy 0.01 Precision level of the continuous variables. Indicates the number of discrete elements the value range is divided by
Details No Enables accuracy specification for every continuous variable
DefaultAccuracy 0.01 Common accuracy value if detailed specification is disabled
AccuracyIncrease 0.2 Accuracy change factor by which accuracy is multiplied when passing to next stage. Provides a means to increase accuracy in subsequent stages. Thus should have a value greater than 0.
Elite 1 Number of elitist designs to be kept (1-5) while moving to the next generation. Option must have an integer value strictly greater than 0 and less than the PopulationSize.
GrayCode No Enables gray code binary implementation. If disabled, fixed point binary representation is used.
MutationRate 0.01 Probability of mutation for each bit. Typically 0.001 to 0.1 and valid range is between 0 and 1.
PopulationSize 60 Size of the population for algorithm to begin with. Option must have integer value greater than or equal to 20.
RevivalChance 0 Chance of revival of lost bits (0.0001 . 0.002). The valid range for the option is between 0 and 1.
Seed Random seed (optional). This option is optional but specifying a valid integer seed value could help generate the same random numbers and thus have same populations between 2 different algorithm runs provided all other parameters stay the same.
WindowSize 3 The worst values of pseudo objective are selected from given number of previous generation. The option must have a positive integer value.
Output
Clusters 1 Number of clusters to be formed by the designs of the last generation and saved to clusters.dat file. Clustering will have better effect if used along with sharing. 1 for regular output (without clustering). Option must have a positive integer value less than PopulationSize.
OutputFilesPath Path where output files are to be generated. Defaults to user's temporary directory if not specified
PrintDetails Yes Enables printing details for each generation
PrintOutFrequency 60 Printing frequency control. Number of evaluations after which data is written to the history.dat file. Option must have a positive integer value.
Penalty Function
MaxCap 7 Initial maximum cap on the penalty of constraint violations. Any penalty which exceeds this maximum cap will be adjusted to the sum of the maximum cap and 20% of the amount of penalty over the value of MaxCap. Option must have a value greater than 0.
PenaltyAdjuster 1 Penalty adjuster which is added to penalty coefficient after every 10 generations. Option must have a value greater than or equal to 0.
PenaltyCoefficient 2 Starting penalty coefficient for constraint violations. Option must have a positive value.
Sharing Strategy
Alpha 1 Alpha parameter defines the nature of sharing function
Share No Enables sharing strategy enhancing the ability of location global minimum
SigmaShare 0.2 Sigma parameter defines the radius of sharing neighborhood. Option must have a value greater than 0.
Termination Criterion
ConvergenceIndex 0.9 Convergence criterion index. If the portion of bits that have identical zeros or ones for all designs is over this index, the algorithm stops. The option must have a value between 0 and 1.
MaxEvaluations 1500 Maximum number of function evaluations per stage. Option must have a positive integer value.
Varying Granularity
MeltingDesigns 40 Number of designs from previous generation to form initial population using melting strategy. Option must have a positive integer value less than the PopulationSize.
Relaxation Percent values of constraints relaxations for varying granularity approach
Stage 1 40 constraint relaxation for stage 1
Stage 2 20 constraint relaxation for stage 2
Stage 3 0 constraint relaxation for stage 3
Stage 4 0 constraint relaxation for stage 4
Stage 5 0 constraint relaxation for stage 5
Stages 3 Number of stages for varying granularity approach. Up to 5 stages allowed. Option must have an integer value between 1 and 5.