Igor S. Peretta

Exterior lighting computer-automated design based on multi-criteria parallel evolutionary algorithm

Multi-objective evolutionary algorithm to computer-automated exterior lighting design.Web client integrated to a cluster of computers to provide lighting design service.Solution to optimize both illumination quality and energy efficiency.Case study solution presents - 37.5% power consumption and +227.3% global uniformity. A proper professional lighting design implies in a continuous search for the best compromise between both low power consumption and better lighting quality. This search converts this design into a hard to solve multi-objective optimization problem. Evolutionary algorithms are widely used to attack that type of hard optimization problems. However, professionals could not benefit from that kind of assistance since evolutionary algorithms have been unexplored by several commercial lighting design computer-aided softwares. This work proposes a system based on evolutionary algorithms which implement a computer-automated exterior lighting design both adequate to irregular shaped areas and able to respect lighting pole positioning constraints. The desired lighting design is constructed using a cluster of computers supported by a web client, turning this application into an efficient and easy tool to reduce project cycles, increase quality of results and decrease calculation times. This ELCAutoD-EA system consists in a proposal for a parallel multi-objective evolutionary algorithm to be executed in a cluster of computers with a Java remote client. User must choose lighting pole heights, allowed lamps and fixtures, as well as the simplified blue print of the area to be illuminated, marking the sub-areas with restrictions to pole positioning. The desired average illuminance must also be informed as well as the accepted tolerance. Based on user informed data, the developed application uses a dynamic representation of variable size as a chromosome and the cluster executes the evolutionary algorithm using the Island model paradigm. Achieved solutions comply with the illumination standards requirements and have a strong commitment to lighting quality and power consumption. In the present case study, the evolved design used 37.5% less power than the reference lighting design provided by a professional and at the same time ensured a 227.3% better global lighting uniformity. A better lighting quality is achieved because the proposed system solves multi-objective optimization problems by avoiding power wastes which are often unclear to a professional lighting engineer in charge of a given project.

Factorial design analysis applied to the performance of parallel evolutionary algorithms

BackgroundParallel computing is a powerful way to reduce computation time and to improve the quality of solutions of evolutionary algorithms (EAs). At first, parallel EAs (PEAs) ran on very expensive and not easily available parallel machines. As multicore processors become ubiquitous, the improved performance available to parallel programs is a great motivation to computationally demanding EAs to turn into parallel programs and exploit the power of multicores. The parallel implementation brings more factors to influence performance and consequently adds more complexity on PEA evaluations. Statistics can help in this task and can guarantee the significance and correct conclusions with minimum tests, provided that the correct design of experiments is applied.MethodsWe show how to guarantee the correct estimation of speedups and how to apply a factorial design on the analysis of PEA performance.ResultsThe performance and the factor effects were not the same for the two benchmark functions studied in this work. The Rastrigin function presented a higher coefficient of variation than the Rosenbrock function, and the factor and interaction effects on the speedup of the parallel genetic algorithm I (PGA-I) were different in both.ConclusionsAs a case study, we evaluate the influence of migration related to parameters on the performance of the parallel evolutionary algorithm solving two benchmark problems executed on a multicore processor. We made a particular effort in carefully applying the statistical concepts in the development of our analysis.

A Statistical Scheme to Support Performance Comparisons between Different Designs of Evolutionary Algorithms

Evolutionary algorithms are stochastic heuristics which can optimize over special functions, known as fitness functions, by manipulating the structure of candidate solutions known as individuals. Multi-objective evolutionary algorithms can deal with many objectives to be optimized, whether concurrent or divergent, ending by returning an optimal frontier, i.e. a set of solutions all defined as Pareto optimal. The idea behind using evolutionary algorithms to perform computer automated design is to be able to formulate a fitness function that, starting from a candidate solution, could reflect the impact the evaluated individual has on those objectives to be optimized. The case study presented in this work is an application for computer automated exterior lighting design which has some concurrent objectives to be optimized: the energy efficiency and the illumination quality. This work investigates four metrics to illumination quality and two metrics for energy efficiency as possible proposals to the fitness function formulation. Eight variations were designed as combinations of pairs from those metrics. To help in the decision process, the statistical hypothesis test known as difference of means is then used to enable comparisons between those variations. This test is performed two by two and three decision matrices is then derived, the ones about global uniformity, mean electrical power, and mean efficiency class index. The concept of Paretos “statistical dominance”, defined in this work and based on statistical evidences, indicates a final decision about which one from the previous designed variations of fitness function is the more appropriated for the presented problem.

A Robust TEO-Based Speech Segmentation Method For Automatic Speech Recognition

Based on the Teager Energy Operator (TEO), the “TEO-based method for Spoken Word Segmentation” (TSWS) is presented and compared with two widely used speech segmentation methods: “Classical”, that uses energy and zero-crossing rate computations, and “Bottom-up”, based on the concepts of adaptive level equalization, energy pulse detection and endpoint ordering. The implemented Automatic Speech Recognition (ASR) system uses Mel-frequency Cepstral Coefficients (MFCC) as the parametric representation of the speech signal, and a standard multilayer feed-forward network (MLP) as the recognizer. A database of 17 different words was used, with a total of 3,519 utterances from 69 different speakers. Two in three of those utterances constituted the training set for the MLP, and one in three, the testing set. The tests were conducted for each of the TSWS, Classical or Bottom-up methods, used in the ASR speech segmentation stage. TSWS has enabled the ASR to achieve 99.0% of success on generalization tests, against 98.6% for Classical and Bottom-up methods. After, a white Gaussian noise was artificially added to the ASR inputs to reach a signal-to-noise ratio of 15dB. The noise presence alters the ASR performances to 96.5%, 93.6%, and 91.4% on generalization tests when using TSWS, Classical and Bottom-up methods, respectively.

Proposal and Preliminary Investigation of a Fitness Function for Partial Differential Models

This work proposes and presents a preliminary investigation of a fitness evaluation scheme supported by a proper genotype representation intended to guide an under development expansion to EASEA/EASEA-CLOUD platforms to evolve partial differential equations as models for a specific system of interest, starting with measures from that system. A simple proof of concept using a dynamic bidirectional surface wave is presented, showing that the proposed fitness evaluation scheme is very promising to enable automate system modelling, even when dealing with up to \(\pm 10\,\%\) noise-added data.

From Measure Data to Evaluation of Models: System Modeling through Custom Galerkin-Jacobi

This work presents a method to evaluate the quality of candidate models for a given observed system in terms of fitness. Taking a candidate model, i.e. a proposed differential equation, this work uses the Galerkin method with a Jacobi/Legendre polynomial basis to approximate solve it. After, this method computes the mean square error between the approximate solution and the measure data. It ends with a relative grade for the fitness of the model to the system to enable comparisons between other possible candidates. The proposed method is intended to aid evolutionary algorithms to evolve fit models to systems based on their measure data.