Reinforced Event-Driven Evolutionary Algorithm Based on Double Deep Q-network.

The real-world optimization task has long been viewed as a noteworthy challenge owing to its enormous search space. To deal with this challenge, evolution algorithm, especially differential evolution algorithm, attracts our attention owing to the excellent robustness. However, for traditional evolution algorithm, how to determine suitable parameters and strategies is a troublesome problem. To deal with the question the reinforced event-driven evolutionary algorithm (REDEA) based on double deep q-network is proposed which embed the double deep q-learning network into differential evolution algorithm with an event-driven controller. To verify the feasibility and superiority of our proposed algorithm, CEC 2013 test suits are utilized and four state-of-arts evolutionary algorithms are involved as the comparisons. The experimental results present that the proposal algorithm obtains comparable capability in most functions.