2017 Fiscal Year Research-status Report
Improving evolutionary algorithms from population structures and interaction networks
| Project/Area Number |
17K12751
|
|---|
| Research Institution | University of Toyama |
|---|
Principal Investigator |
高 尚策 富山大学, 大学院理工学研究部(工学), 准教授 (60734572)
|
|---|
| Project Period (FY) |
2017-04-01 – 2019-03-31
|
| Keywords | Intelligent algorithm / Population structure / Complex network / Differential evolution / Optimization |
|---|
| Outline of Annual Research Achievements |
In our research, we concentrate our attention on the population which is the common component in all EAs. The population structure and evolutionary dynamics were systematically investigated. The extraction of the generic characteristics of the information interaction network constructed by the population was studied, and the systematical generation of effective search algorithms from the aspect of population structure was designed.
1. We used node degrees to characterize the population interaction network from the view of complex network; 2. We studied population structures affect the information flux in the interaction network; 3. The mechanisms and results of population structures which affect the search performance in terms of solution precision, convergence and population diversity were studied; 4. We have found two rules of population structure which is the most robust (i.e. problem-independent) when solving the optimization problem with single objectives.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
As a result of this research, we have published 10 journal papers and 5 conference papers. Especially, we observed that a power law distribution exists in brain storm optimization (BSO) algorithm and a Poisson law can be derived from population interaction network in differential evolution algorithm:
1. We proposed a population interaction network (PIN) to investigate the relationship constituted by populations. The experimental results demonstrate the CDF meets cumulative Poisson distribution.
2. PIN was used to construct the relationship among individuals in BSO. The experimental results indicated the frequency of average degree of BSO meets a power law distribution in the functions with low dimension, which shows the best performance of algorithm among three kinds of dimensions.
|
| Strategy for Future Research Activity |
In the following research year, we plan to use the population interaction network (PIN) to establish the relationship between other EAs and complex network (CN). Through population dynamic analysis and statistical confidence test via PIN, the topological structure properties (such as degree distribution) in CN can be used to study the issues (population diversity, etc.) in EAs. Theoretical analysis and application verification are also carried out. Furthermore, two key scientific factors will be studied from three aspects: statistics, structures and abstraction.
|
| Causes of Carryover |
The reason of the incurring amount to be used next fiscal year is mainly because the payments of some accepted or conditional accepted papers have not been finished. And we plan to use it as additional fees of personnel expenditure and remuneration.
|
