Analysis of hyper-heuristic performance in different dynamic environments

Analysis of global information sharing in hyper-heuristics for different dynamic environments

“…This study uses this comprehensive dynamic environment classification to determine if different kinds of hyperheuristics perform differently across various types of dynamic optimisation problems. Van der Stockt and Engelbrecht found that a random selection hyper-heuristic can outperform the individual constituent algorithms in certain types of dynamic environments [14]. This paper extends the previous study by investigating three different hyper-heuristics across different types of dynamic environments.…”

“…Van der Stockt and Engelbrecht's initial study did not share global state information [14]. This paper extends Van der Stockt and Engelbrecht's previous study by analysing HMHH performance when each managed heuristic has access to the information of all the other heuristics in the pool.…”

“…The aim is to obtain improved performance when compared to running those same heuristics in isolation, either through better overall accuracy, faster convergence, or both after landscape changes occur. Van der Stockt and Engelbrecht demonstrated that increased performance is possible using hyper-heuristics in dynamic environments [14].…”

“…Simple Gaussian mutation operators were used as heuristics, the moving peaks benchmark was used, and only spatial and temporal changes were considered. • Van der Stockt and Engelbrecht investigated a simple random hyper-heuristic to manage a pool of metaheuristics that are specialised to solving dynamic optimisation problems [14]. The HMHH (without global information sharing) as outlined above was used.…”

“…The intuitive expectation is that hyper-heuristics should be well-suited to solve dynamic optimisation problems because different heuristic search methods can be brought to bear at different times during the search. Recently a number of studies have investigated this expectation [9] [10] [11] [12] [13] [14].…”