Evolutionary algorithm with a novel insertion operator for optimising noisy functions

As more complex engineering optimisation problems are being tackled, optimisation algorithms are being stretched to their limits. The simulations are often subject to noise and uncertainties, leading to noisy objective functions and variable evaluation times. With a noisy objective, it is often very difficult to identify the best solutions reliably as the noise can cause even an optimum value to appear `good´ rather than `excellent´. Bad solutions are often much easier to spot as it takes a lot of noise to make them appear `good´. Thus in a noisy environment, a strategy of replacing bad solutions may have an advantage over selecting excellent ones. A new insertion process has been developed that allows the insertion strategy to be tuned smoothly between `greedy´ fitness based insertion and uniform random insertion. This new insertion function allows both the evolutionary processes of selection and insertion to be adjusted to suit the level of noise and complexity in the objective calculations. This paper demonstrates that the insertion and selection operators can be tuned to suit the level of noise in the objective to maintain maximum algorithm efficiency and solution accuracy. Experiments have shown that for some noisy problems, the insertion process must dominate the selection operator for maximum efficiency, but the selection process has a significant effect on the accuracy of the final solution