Oscillation-based Linear Dynamic Sampling Allocation for Noisy Multiobjective Optimization Problems

Abstract

The physical quantities in real-life can be measured by using devices called sensors. These devices are not perfect/ideal devices which have the properties called sensitivity and resolution. As the sensors became more sensible with a higher resolution the current/real physical data can be obtained. However even by using the best possible sensor, the sensor still under the measurement noise. Therefore, it is natural to have noise in engineering problems. To solve these engineering problems the noise should be considered in the calculations. In Multiobjective optimization this noise can be added to the objectives and called noise Multiobjective optimization problems. To solve these problems the most common method is called re-sampling which is the calculation the objectives many times and taking the average of their values. The dynamic re-sampling is a method for efficient with respect to the computational source. In this research a new dynamic re-sampling method is proposed and named as oscillation-based linear dynamic sampling method. This method is integrated into four different Multiobjective optimization algorithms and applied to eight benchmark problems. The results showed that the proposed method gives acceptable results with relatively small number of additional function evaluation.