@PhdThesis{dbt_mods_00002890,
  author = 	{Meyer Dr. - Ing., Dany},
  title = 	{Modellbasierte Mehrzieloptimierung mit neuronalen Netzen und Evolutionsstrategien},
  year = 	{2005},
  month = 	{Apr},
  day = 	{15},
  keywords = 	{Neuronale Netze; Evolution{\"a}re Algorithmen; Evolutionsstrategien; multikriterielle; modellbasierte Optimierung; Mehrzieloptimierung; Vektoroptimierung},
  abstract = 	{Model-based Multi-objective Optimization with Neural Networks and Evolution 
Strategies

Abstract

Today, tasks of optimization are not excluded from any part of the modern 
engineering. Ever more frequently engineers must cope with complex optimization 
problems with conflicting goals as well as a large number of various 
constraints. Additionally, there is the claim to incorporate human expert 
knowledge in an easy and transparent manner within the solutions. Often the 
dependencies within the parameters of the process which is to be optimized are 
mathematically no more or only very imperfectly formulatable, are present in the 
form of rule knowledge, or example situations and must therefore be estimated by 
suitable models. These requirements present complex challenges for the 
developers of modern concepts and optimization methods, which are in most cases 
no longer solvable with methods of classical mathematics alone. Rather, they 
require the additional employment of adaptive methods, which lead by using 
synergies between the classical and nature-analog procedures for the development 
of efficient hybrid systems.
In this work a multitier system for model-based, multi-objective optimization is 
presented, which consists of the data driven process modeling for calculation of 
objectives and constraints, their multi-objective optimization as well as an 
interactive Decision Making System. The uniqueness of the presented approach is 
the development of modeling and interpolation of the generated pareto optimal 
solutions and their corresponding objectives after the optimization by neural 
networks. In this way the approach allows to perform an interpolation access 
within the pareto set as well as the extraction of knowledge between the process 
variables near the pareto set and pareto front. Besides the representation of a 
practical-suited methodology, extensions in the theory of evolutionary 
algorithms in the form of learning the evolution direction during the 
optimization represents a further emphasis. The additional combination with a 
gradient-based optimization algorithm makes the approach a multi-hybrid system, 
which is characterized by very good convergence characteristics and a high 
quality of the generated solutions.
As an example of the industrial application of the presented approach, a system 
for model-based, multi-objective recipe optimization in the animal fodder 
industry is described. The aim of the work is the development of an adaptive, 
multi-hybrid multi-objective evolutionary algorithm which exhibits its 
superiority by efficiently using synergies between different natur-analog and 
mathematical methods as well as the presentation of a practical methodology for 
engineers to optimize the production processes. This includes a more efficient, 
powerful design of experiments, process modeling and multi-objective 
optimization.},
  url = 	{https://www.db-thueringen.de/receive/dbt_mods_00002890},
  url = 	{http://uri.gbv.de/document/gvk:ppn:483812153},
  file = 	{:https://www.db-thueringen.de/servlets/MCRZipServlet/dbt_derivate_00004625:TYPE},
  language = 	{de}
}