Innovative Meta-heuristics for Solving Non Linear Inverse Problems in Industrial and Information Engineering

Abstract

In the world of technology, many industrial operations such as design of efficient devices, or planning production in a big factory, require optimization approach and the solution of inverse problems[chapter 1]. In this contest, in the last 20 years, the heuristic methods had a primary role considering their capabilities to find out solutions in all those cases in which a lot of computation time is requested. The present thesis work is mainly based on swarm algorithms, and on their capabilities to achieve global optima without remain trapped into local minima. In particular, in this work we treat high hybridization and integration among the different capabilities in exploitation and exploration, expressed by 3 optimization algorithms which are: Particle Swarm Optimization (PSO), Flock of Starlings Optimization (FSO), Bacterial Chemotaxis Optimization (BCA). The research of high hybridization among different heuristics led to implement a new metaheuristic which has been called MeTEO (Metric Topological Evolutionary Optimization). MeTEO exploits the synergy among the three algorithms already mentioned above. Moreover, in MeTEO a further method called Fitness Modification (FM) has been used. As will be shown, the FM enhance the exploration properties of MeTEO together with benefits in the parallelization. The first problem encountered making a metaheuristics composed of three complex algorithms is the computation time required. For this reason, the thesis work has been focused also in the analysis and synthesis of a parallel structures for supporting calculus. In this context, two different approaches have been studied: 1)the algorithm-based and 2) the fitness-based. Moreover, in order to extend the exploration capability of FSO problems with discrete variable as well, a binary version of FSO has been implemented [chapter 2]. MeTEO has been validated on benchmarks and on inverse problems. The benchmarks used are called hard benchmarks, because they show a structure without preferential tendency towards a particular point, and local minima with depth value, some monomodal, with one global minimum, and multimodal, with many equivalent minima. Afterwards a list of real inverse and optimization problems are proposed: the parameters identifications of Jiles-Atherton parameters, the efficiency improvement of Travelling Wave Tube (TWT) device, taking in account the geometry, the magnetic focusing field, and the voltage of a Multistage Compressed Collector, the harmonic detection in distorted waveforms. The simulation has been made with MATLAB, and with the help of a FEM simulator called COLLGUN. All results have been compared with those from other algorithms such as random walk, and the also from the use of a single heuristics which MeTEO exploits [chapter 3]. In the Chapter 4 of this thesis the point of view changes toward the hardware, whereas all the discussion done in the previous three chapters were focused on the improvement of the optimization process preformed by numerical algorithms (Software). In fact, we present a method for translating a numerical swarm based algorithm into an electric circuit, that is able to reproduce by mean of voltages and currents the same trajectories shown by the numerical swarm-based algorithms. A circuit, called swarm circuit, has been implemented with Simulink after to have deduced the mathematical relations between the numerical algorithms and their translation into a dynamic system. The swarm circuit has been tested with hard benchmarks and with two inverse problems. The swarm circuit open the road towards a real time optimization, argument that is difficult to be addressed with software approaches.