DEPSO Algorithms: Project Portal
URL: http://www.adaptivebox.net/doi/DEPSO
(Current Version: V1.0.002)

Basic Description What's New Problem to be solved Setting Parameters Output Information References

DEPSO, or called DEPS, is an algorithm for (constrained) numerical optimization problem (NOP), which hybridizes the advantages of both Particle Swarm Optimization (PSO) and Differential Evolution (DE).

License information: DEPSO is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License 3.0.

System Requirements: DEPSO is a platform-independent software developed by JAVA version 1.4 or above.

Command line (examples): $ java DEPSO Problem=<Problem_Name> [NAME=VALUE] ...

What's New

Version: V1.0.002 [download]:

  • the stochastic selection of PS and DE operators is introduced, in which the selection probability for PS is switchP (the old versions simply choose PS and DE in odd and even cycles, respectively).
    • // There are three typical settings: 1) the stochastic selection strategy is similar to the old one if switchP=0.5; 2) it is the pure PSO if switchP=1.0; 3) it is the pure DE if switchP=0.0.
  • For PS operator, a chaos factor (CL) [4] is introduced, which give the probability of introducing a purely random move in each dimension of the search space.
    • // The new PS operator is identical to the the old one if CL=0.
  • Newly introduced parameters: switchP, CL

    • Version: V1.0.001 [download]:

  • For boundary-handling, the cycled version [3] instead of the periodic version [1] is considered, so that all new solutions are generated within the original search space, as well as the agents are searching within a virtually infinite space. In addition, the limitation of maximal velocity is no longer required.
  • The adaptive constraints relaxing (ACR) rule [2] might tackle the problem with equality constraints more efficiently than the basic constraint-handling (BCH) rule does.
  • Newly introduced parameters: isACR.

    • Version: V1.0.000 [download]:

  • It implements the original DEPSO algorithm [1].
  • Setting parameters: Problem, N, T, Tout, FACTOR, CR, c1, c2, weight.

    • Problem to be solved

    The problem to be solved is (constrained) numerical optimization problem (NOP), or called the nonlinear programming problem.

    Tips: 1) all the variable bounds must be specified, since optimal solution(s) might situate at anywhere; 2) it had better to avoid using any equality constraints, at least any unavoidable constraint should be relaxed by a small tolerance value (e.g., ε=1E-4 for problem.constrained.Michalewicz_G3); and 3) problem.ProblemEncoder andproblem.UnconstrainedProblemEncoder are the parental classes of all constrained (e.g., problem.constrained.Michalewicz_G1) and unconstrained (e.g., problem.unconstrained.GoldsteinPrice) problems, respectively.

    Implemented problem instances: please download from the up-to-date list of source files, which will be situated in the directories: 1) problem/constrained, and 2) problem/unconstrained.

    Setting parameters [NAME=VALUE]
    		 

    NAME    VALUE_type   Range      Default_Value   Description
Problem  String      *		<Problem_Name>  The problem to be solved
//For example: problem.constrained.Michalewicz_G2 is the default value
-----------------------------------------------------------------------------------------

N        integer     >5         70 		The number of agents
T        integer     >1         2000		The maximum learning cycles
//The total number of evaluation times is about N*T

isACR    boolean                false           Constraint-handling: BCH(false), ACR(true)
//Basic constraint-handling (BCH) rule or adaptive constraints relaxing (ACR) rule

Tout     integer     >0         100		The output interval (not important)
//The program outputs runtime information of the best solution every "Tout" cycles.

FACTOR   real        (0, 1.2]   0.5  	        DE: scale constant
CR       real        [0, 1]     0.9  	        DE: crossover constant
//The parameters of DE operator, there are two suggested settings for DE:
// 1) FACTOR=0.5, CR=0.9; 2) FACTOR=0.5, CR=0.1

c1       real        [0, 2]     1.494           PSO: learning factor for pbest
c2       real        [0, 2]     1.494           PSO: learning factor for gbest
w        real        [0, 1]     0.729           PSO: inertia weight
CL       real        [0, 0.01]  0.000           PSO: Chaos factor (see Ref.[4])
//The parameters of PSO operator, there are two suggested settings for PSO:
// 1) c1=c2=1.494, weight=0.729, CL=0; 2) c1=c2=2.0, weight=0.4, CL=0.001

switchP  real        [0, 1]     0.5             DEPSO: selection probability

    Output Information

    [Parsing information]: provide the parsing information for all input parameters.

    [Setting information]: show the information of all setting parameters for the algorithm.

    [Runtime information]: The program outputs runtime information, i.e., the evaluation values <Vcon, Vopt> of the best solution, at every "Tout" cycles.
    //Vopt: the value of objective function; Vcon: the weighted constraint violation value (≥0), which is not outputted if Vcon≡0 since there is no violation

    [Summary information]: At the end, it outputs the input variables, response values, and evaluation values <Vcon, Vopt> of the best solution.

    References

    [1] Wen-Jun Zhang, Xiao-Feng Xie. DEPSO: hybrid particle swarm with differential evolution operator. IEEE International Conference on Systems, Man, and Cybernetics (SMCC), Washington, DC, USA, 2003: 3816-3821. [DOI]

    [2] Xiao-Feng Xie, Wen-Jun Zhang, De-Chun Bi. Handling equality constraints by adaptive relaxing rule for swarm algorithms. Congress on Evolutionary Computation (CEC), Portland, OR, USA, 2004: 2012-2016. [DOI]

    [3] Xiao-Feng Xie, Jiming Liu. A compact multiagent system based on autonomy oriented computing, IEEE/WIC/ACM International Conference on Intelligent Agent Technology (IAT), Compiégne, France, 2005: 38-44. [DOI]

    [4] Xiao-Feng Xie, Wen-Jun Zhang, Zhi-Lian Yang. A dissipative particle swarm optimization. Congress on Evolutionary Computation (CEC), Honolulu, HI, USA, 2002: 1456-1461. [DOI]

    Return to homepage

    Maintained by AdaptiveBox StUdIo, under a Creative Commons Attribution 3.0 License.