Load balance aware distributed differential evolution for computationally expensive optimization problems
Ning Ma, Xiaofang Liu, Zhi‐Hui Zhan, Jinghui Zhong, Jun Zhang · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
Computationally expensive problem challenges the application of evolutionary algorithms (EAs) due to the long runtime. Distributed EAs on distributed resources for calculating the individual fitness value in paralllel is a promising method to reduce runtime. A crucial issue in distributed EAs is how to scheduling the individuals to the distributed resources. Different resources are often with different load and the resource with slow computation ability often limits the parallel speed. To improve the performence, the load information of each resource is considered and used for resource allocation strategy in this paper. We proposed a distributed differential evolution (DDE) algorithm with a load balance strategy to efficiently utilize the concurrent computational resource for power electronic circuit design, which is a computationally expensive optimization problem. This way, the topology related to the individuals and the resources will be adaptively changed. Experiments on distributed resources are carried out to evaluate the effect of the load balance based allocation strategy. The results indicate that the proposed load balance strategy is able to significantly reduce the runtime.