Strategies for enhanced performance of single & multi-objective optimization using evolutionary algorithms: applications to dynamics optimization

Show simple item record

dc.contributor.advisor Padhiyar, Nitin
dc.contributor.author Patel, Narendra Madhavlal
dc.date.accessioned 2017-04-01T14:10:53Z
dc.date.available 2017-04-01T14:10:53Z
dc.date.issued 2016
dc.identifier.citation Patel, N. M. (2016). Strategies for enhanced performance of single & multi-objective optimization using evolutionary algorithms: application to dynamics optimization. Indian Institute of Technology Gandhinagar, pp. 204. (Acc No: T000189) en_US
dc.identifier.uri https://repository.iitgn.ac.in/handle/123456789/2813
dc.description.abstract Evolutionary Algorithm (EAs) are often criticized for their large computational cost for solving practical optimization problems. Moreover they are naturally designed for unconstrained problems and hence require an additional mechanism for constraint handling. On the other hand Voz complex method is a gradient free local search optimization technique having good convergence and constraint handling capabilities. Furthermore, the Box-Complex method being the multi-start numerical optimization technique, its hybridization with the EAs is quite straight forward. We explore this hybridization for enhancing the convergence and constraint handling capabilities of the EAs. We add one or more population members created by Box-Complex method using the current population and replace and replace the equal number of worst population members violating the constraints are projected through the centroid of a geometric complex made of a few feasible points using Box-Complex method for constraint handling. The success measure of the single objective EAs is its capability to converge to the global optimum. On the other hand, there ae three measures for multi-objective optimization (MOO), namely convergence to the global pareto, the uniform spread of the solutions, and the coverage of the pareto front. Box-Complex is guided based in the objective (fitness) function value. This makes its hybridization with SOO EAs quite straightforward. However, such hybridization is not straightforward for the MOO problems since there are multiple conflicting fitness criteria in MOO. This motivated us to develop a novel sorting mechanism for MOO EAs by discretizing the objective space into an m-dimensional mesh, where m is the number of objectives. In this mesh sorting, the overall fitness is defined based on the location of the population member in the mesh. Furthermore, the additional MOO criteria can easily be incorporated with the novel mesh sort framework. The above, mentioned proposed strategies are quite generic and can be implemented with any population based EA. However, in the current work, we use Genetic Algorithm (GA), Cuckoo Search (CS), and Differential Evolution (DE) to demonstrate their performance. All the proposed strategies are first tested using the benchmark test problems. The strategies have been further validated using various engineering applications. Dynamic optimization (DO) is a class of optimization that optimizes the temporal profiles of the process inputs satisfying the process dynamics. Solving such DO problems is computationally expensive and quite challenging compared to the static optimization problems. Hence, we also validate the proposed strategies in the EAs using single and multi-objective DO problems. With increased computational advancements, there is a significant scope of utilizing the multi-objective optimization tool for enhancing the process efficiency. We in this work also propose novel multi-objective dynamic optimization problem formulations for a class of fed-batch bio-reactors. We solve four bi-objective problems and a three-objective problem for a n application of the secreted protein production in a fed-batch bio-reactor. The four objectives considered in this study are maximization of productivity, maximization of yield, minimization of fed-batch operation time, and minimization of the endpoint substrate concentration. en_US
dc.description.statementofresponsibility by Narendra M. Patel
dc.format.extent 204 p.; ill.; 30 cm+.
dc.language.iso en_US en_US
dc.publisher Indian Institute of Technology Gandhinagar en_US
dc.title Strategies for enhanced performance of single & multi-objective optimization using evolutionary algorithms: applications to dynamics optimization en_US
dc.type Thesis en_US
dc.contributor.department Chemical Engineering
dc.description.degree Ph.D.


Files in this item

Files Size Format View

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record

Search Digital Repository


Browse

My Account