Using Neural Networks and Diversifying Differential Evolution for Dynamic Optimisation

Dynamic optimisation occurs in a variety of real-world problems. To tackle\nthese problems, evolutionary algorithms have been extensively used due to their\neffectiveness and minimum design effort. However, for dynamic problems, extra\nmechanisms are required on top of standard evolutionary algorithms. Among them,\ndiversity mechanisms have proven to be competitive in handling dynamism, and\nrecently, the use of neural networks have become popular for this purpose.\nConsidering the complexity of using neural networks in the process compared to\nsimple diversity mechanisms, we investigate whether they are competitive and\nthe possibility of integrating them to improve the results. However, for a fair\ncomparison, we need to consider the same time budget for each algorithm. Thus,\ninstead of the usual number of fitness evaluations as the measure for the\navailable time between changes, we use wall clock timing. The results show the\nsignificance of the improvement when integrating the neural network and\ndiversity mechanisms depends on the type and the frequency of changes.\nMoreover, we observe that for differential evolution, having a proper diversity\nin population when using neural networks plays a key role in the neural\nnetwork's ability to improve the results.\n