Leonardo de Mello Honório

A gradient-based artificial immune system applied to optimal power flow problems

Mathematically, an optimal power flow (OPF) is in general a non-linear, non-convex and large-scale problem with both continuous and discrete control variables. This paper approaches the OPF problem using a modified Artificial Immune System (AIS). The AIS optimization methodology uses, among others, two major immunological principles: hypermutation, which is responsible for local search, and receptor edition to explore different areas in the solution space. The proposed method enhances the original AIS by combining it with a gradient vector. This concept is used to provide valuable information during the hypermutation process, decreasing the number of generations and clones, and, consequently, speeding up the convergence process while reducing the computational time. Two applications illustrate the performance of the proposed method.

Artificial Immune System Applied to the Multi-stage Transmission Expansion Planning

Transmission expansion planning (TEP) is a complex optimization task to ensure that the power system will meet the forecasted demand and the reliability criterion, along the planning horizon, while minimizing investment, operational, and interruption costs. Metaheuristic methods have demonstrated the potential to find good feasible solutions, but not necessarily optimal. These methods can provide high quality solutions, within an acceptable CPU time, even for large-scale problems. This paper presents an optimization tool based on the Artificial Immune System used to solve the TEP problem. The proposed methodology includes the search for the least cost solution, bearing in mind investments and ohmic transmission losses. The multi-stage nature of the TEP will be also taken into consideration. Case studies on a small test system and on a real sub-transmission network are presented and discussed.

Artificial Immune Systems and Differential Evolution Based Approaches Applied to Multi-Stage Transmission Expansion Planning

Transmission expansion planning (TEP) is a complex optimization task to ensure that the power system will meet the demand in an adequate quality level to customers along the planning horizon, while minimizing investment, operational, and interruption costs. Optimization approaches based on metaheuristics have demonstrated a good potential to find high quality solutions. Their success is related to the ability to avoid local optima by exploring the basic structure of each problem. Numerous advantages can be linked to these tools: a simple software complexity, an ability to mix integer and non-integer variables, and a faster time-response. This paper presents a performance comparison between two optimization tools based on Artificial Immune Systems and Differential Evolution to solve the multi-stage TEP problem. The proposed methodology includes the search for the least cost solution, bearing in mind investments and operational costs related to ohmic transmission losses. The multi-stage nature of the TEP is also taken into consideration. Case studies on a small test system and on a real sub-transmission network are presented and discussed.

Dynamic Polymorphic Agents Scheduling and Execution Using Artificial Immune Systems

When a set of heterogeneous agents is considered to solve different kinds of problems, it is very challenging to specify the necessary number of elements, which functionally of each one will be used and the schedule of these actions in order to solve these problems. To deal with scenarios like this, the present article suggests an innovation at the Intelligent Agent Theory, a new concept called Dynamic Polymorphic Agent (DPA). This approach implies on the dynamic generation of one agent, built from the cooperation of existing agents and specific to fulfill the demanding task. To create this new entity, a monitor identifies and reads information regarding the functionalities of available agents present in the scene and, when a new problem is presented, it generates a task list to solve it. This list and the agents whose functionalities are necessary to solve the problem generate the new polymorphic agent. To fulfill this approach, two major paradigms are used: Aspect-Oriented Program (AOP) and Artificial Immune System (AIS).

A multiple kernel classification approach based on a Quadratic Successive Geometric Segmentation methodology with a fault diagnosis case

This work presents a new approach for solving classification and learning problems. The Successive Geometric Segmentation technique is applied to encapsulate large datasets by using a series of Oriented Bounding Hyper Box (OBHBs). Each OBHB is obtained through linear separation analysis and each one represents a specific region in a patterns solution space. Also, each OBHB can be seen as a data abstraction layer and be considered as an individual Kernel. Thus, it is possible by applying a quadratic discriminant function, to assemble a set of nonlinear surfaces separating each desirable pattern. This approach allows working with large datasets using high speed linear analysis tools and yet providing a very accurate non-linear classifier as final result. The methodology was tested using the UCI Machine Learning repository and a Power Transformer Fault Diagnosis real scenario problem. The results were compared with different approaches provided by literature and, finally, the potential and further applications of the methodology were also discussed.

Learning Fuzzy Systems by a Co-Evolutionary Artificial-Immune-Based Algorithm

To create a Fuzzy System from a numerical data, it is necessary to generate rules and memberships representing the analyzed set. This goal demands to break the problem into two parts: one responsible for learning the rules and another responsible for optimizing the memberships. This paper uses a Gradient-based Artificial Immune System with a different population for each of these parts. By simultaneously co-evolving these two populations, it is possible to exchange information between them enhancing the fitness of the final generated system. To demonstrate this approach, a fuzzy system for autonomous vehicle maneuvering was developed by observing a human driver.

Concepts of Aspect-Oriented Modeling Applied to Optimal Power Flow Problems

Optimization of complex systems demands advanced methods that are implemented in specialized software. Multiple combinations of optimization methods, objective functions, and constraints further complicate the problem of developing this software, making it hard to create, maintain, and evolve. To overcome this problem, this paper presents a new development methodology based on ideas of Aspect-Oriented Programming (AOP) applied to optimal Power Flow Problems. This new methodology supports a clean separation of concerns, and keeps dependencies to a minimum. The optimization method is self- contained and completely independent from the rest of the system; for each optimization scenario, the solution binds the optimization with the concrete problem at run-time. This approach improves the ability to deal with several different objective functions and constraints, providing flexibility, maintainability, and usability to the development and evolution effort without degradation of the computational time. To evaluate this model, it is compared with traditional OOP paradigm using several software metrics.

Persistently-exciting signal generation for Optimal Parameter Estimation of constrained nonlinear dynamical systems

This work presents a novel methodology for Sub-Optimal Excitation Signal Generation and Optimal Parameter Estimation of constrained nonlinear systems. It is proposed that the evaluation of each signal must also account for the difference between real and estimated system parameters. However, this metric is not directly obtained once the real parameter values are not known. The alternative presented here is to adopt the hypothesis that, if a system can be approximated by a white box model, this model can be used as a benchmark to indicate the impact of a signal over the parametric estimation. In this way, the proposed method uses a dual layer optimization methodology: (i) Inner Level; For a given excitation signal a nonlinear optimization method searches for the optimal set of parameters that minimizes the error between the outputs of the optimized and benchmark models. (ii) At the outer level, a metaheuristic optimization method is responsible for constructing the best excitation signal, considering the fitness coming from the inner level, the quadratic difference between its parameters and the cost related to the time and space required to execute the experiment.

Otimização do Processo de Construção de Redes Neurais Artificiais com Método de Segmentações Geométricas Sucessivas Utilizando Projeções de Estimador de Densidade para Quebra em Hipervolumes

This work presents a new technique for otimization of the process of artificial neural networks assembling and training named Successive Geometric Segmentation Method (SGSM). The SGSM groups the data of each class into hyperboxes (HB) aligned in accordance with the largest axis of its points distribution. If the HB are linearly separable, a separating hyperplane may be identified resulting a neuron. If it is not, the data is divided into smaller classes for new HB. In this case, the chosen technique uses the probability distribution of the data in a hyperbox. Resumo—Este trabalho apresenta uma técnica a fim de otimizar o processo de construção e treinamento de redes neurais no Método de Segmentações Geométricas Sucessivas (MSGS). O MSGS agrupa os dados de cada classe em hipercaixas (HC) onde cada caixa é alinhada de acordo com os eixos de maior distribuição de pontos. Sendo as caixas linearmente separáveis, um hiperplano de separação é identificado originando um neurônio. Caso não seja posśıvel a separação por um único hiperplano, os dados são divididos em conjuntos menores para obter novas HC. Neste caso, a técnica de divisão estima a densidade de probabilidade dos pontos na hipercaixa para escolher o ponto de corte.

Diagnóstico de faltas em transformadores de potência através do método de segmentações geométricas sucessivas

Este trabalho apresenta um novo metodo classificador para o diagnostico de faltas incipientes em transformadores de potencia com base em levantamento de dados pela tecnica de analise de gases dissolvidos. O classificador fundamenta-se na construcao e treinamento de redes neurais artificiais a partir de segmentacoes geometricas sucessivas. Essa abordagem permite gerar tanto a topologia quanto o peso de cada neuronio sem a necessidade da especificacao dos parâmetros da rede. O classificador de padroes foi testado utilizando-se conjuntos de dados de faltas em transformadores de potencia disponiveis na literatura e os resultados dos testes realizados indicam altas taxas de acerto.