Executive Summary : | Multi-objective problems have less than or equal to three objective functions, while many-objective optimization problems have four or more. Finding the optimal solutions to the many-objective optimization problems is considered challenging by the optimization research community. Most of the current multi-objective evolutionary algorithms (MOEAs) perform poorly with the increase in the no. of objectives and may not be suitable for solving many-objective optimization problems. Furthermore, simultaneous optimization of many objectives containing large no. of variables lead to a huge Pareto front. With so many Pareto optimum solutions available, the decision-maker (DM) faces difficulty in choosing a particular solution. By using a pre-defined set of rules, the pruning techniques aid the DM in selecting a particular solution by identifying a subset of Pareto optimal solutions. However, many pruning techniques may end up being ineffective due to a large number of design or decision variables and several objectives present in the many-objective optimization problems. So, this gap provides researchers with an opportunity to develop new pruning techniques that can reduce the no. of objectives at a relatively early stage of the optimization procedure. Hence, novel techniques will be developed in this work employing a novel dimensionality reduction technique to lower the no. of objectives either before and/or during the optimization procedure. The use of supervised filter models based on diversity or redundancy will be researched. In some diversity-based pruning techniques, hypervolume was used as an indicator for the pruning instruction. However, no study other than the hypervolume-based approach has yet been done that explicitly employs diversity optimization techniques with the goal of reducing the Pareto front. This creates lot of scope to develop and use new pruning techniques. A comparison study will be conducted to demonstrate the superiority of the proposed technique over the current diversity-based techniques. An ensemble pruning technique that yields a compromise pruned set will be developed. This technique will yield better outcomes than a single model. Also, a diversity-based pruning technique will be combined with an efficiency-based pruning technique. This opens up another possibility, enabling the DM to select the best knee from a set of competing solutions using multi-attribute decision-making (MADM) methods. Pareto optimality under uncertainty will also be considered and novel stochastic pruning methods will be developed. New performance indicators to assess the stochastic pruned set will be developed and the viability of combining the current indicators will be examined. The new pruning techniques will be used for many-objective design optimization of selected thermal systems and devices like different types of heat exchangers, heat sinks, heat pumps, rotary regenerators, cooling towers, and thermo-electric coolers. |