In the present paper we propose a hybrid approach of Differential Evolution algorithm in noisy optimization problems. The classical algorithm is combined with a reinforcement learning technique and polynomial extrapolation to allocate intelligently the available computational budget to promising candidate solutions, guiding the population to the global optimum by reducing the effect of noise. Experimental tests are performed on five established benchmark problems, with varying dimensionality and noise levels, and the results are statistically compared with an approach based on Particle Swarm Optimization method. Useful conclusions are derived and questions that require further investigation on both methods arise.