Stochastic Search Technique Research Articles

Design optimization with advanced genetic search strategies

The present paper describes the capabilities of a modern design optimization tool based on the method of genetic search. This stochastic search technique offers a significantly increased probability of locating the global optimum in a design space with multiple relative optima. The program includes an advanced search technique referred to as directed crossover wherein bit positions on the design strings that offer a higher gain during crossover are assigned higher probabilities of selection as crossover sites. This optimization code also includes a multistage genetic search plan that is useful in problems where the design space is large. Multistage search involves successive refinement in the precision with which design variables are represented in the genetic search process. Also included in this program is a newly developed cluster identification technique that automatically determines the center location and the radius of a hypersphere representing a relative-optimum containing region. Cluster information serves to define accurate parameters required for other advanced techniques such as sharing function implementation and mating restrictions.

Trefoil knotting revealed by molecular dynamics simulations of supercoiled DNA.

Computer simulations of the supercoiling of DNA, largely limited to stochastic search techniques, can offer important information to complement analytical models and experimental data. Through association of an energy function, minimum-energy supercoiled conformations, fluctuations about these states, and interconversions among forms may be sought. In theory, the observation of such large-scale conformational changes is possible, but modeling and numerical considerations limit the picture obtained in practice. A new computational approach is reported that combines an idealized elastic energy model, a compact B-spline representation of circular duplex DNA, and deterministic minimization and molecular dynamics algorithms. A trefoil knotting result, made possible by a large time-step dynamics scheme, is described. The simulated strand passage supports and details a supercoiled-directed knotting mechanism. This process may be associated with collective bending and twisting motions involved in supercoiling propagation and interwound branching. The results also demonstrate the potential effectiveness of the Langevin/implicit-Euler dynamics scheme for studying biomolecular folding and reactions over biologically interesting time scales.

Optimizing electromagnetic devices combining direct search methods with simulated annealing

Problems of optimization in computational electromagnetics are discussed with particular reference to the determination of the global minimum of practical objective functions. An optimizing environment is described which uses integral methods for calculating user-defined objective functions. Results are presented which combine stochastic and direct search techniques to find a satisfactory design optimum of practical problems. >

Stochastic approximation methods for identification and control--A survey

Stochastic search techniques have been the essential part for most identification and self-organizing or learning control algorithms for stochastic systems. Stochastic approximation search algorithms have been very popular among the researchers in these areas because of their simplicity of implementation, convergence properties, as well as intuitive appeal to the investigator. This paper presents an exposition of the stochastic approximation algorithms and their application to various parameter identification and self-organizing control algorithms.