Modeling Reinforcement Learning and Meta-Level Control for Improving Performance of Communication in Management of Air Traffic

This paper describes a model of intelligent computing that assists decision making in the exchange of messages in a distributed system. The model acts as an additional layer that uses metadata in their decisions. It can be considered innovative because it uses reinforcement learning suited to the characteristics of a stochastic environment, concerned with the speed and quality of decision-making. We propose three strategies for learning: heuristic initial epsilon adaptive and heuristic based on performance. These are combined with reinforcement learning algorithms: Q-learning and SARSA. Case studies evaluate the performance and quality of learning about the strategies proposed.


14. BUENO BOGES SOUZA, 13/06/2008

Methodology balancing smart about multi-streams for use in managing air traffic flow

The Balancing Module Flow (MBF) is proposed to support the system in operation in the First Integrated Center for Air Defense and Air Traffic Control (CINDACTA I) and improve the management process applied by the controllers in this center. Using flow maximization techniques adapted from graph theory, the MBF was developed as a model of analysis that determines the separation time between the departures from terminals contained in the Flight Information Region of Brasilia (FIR-BS) and distributes off flow along the controlled airspace. The goal is to prevent or reduce congestion in the various sectors of the FIR-BS.

With the help of other system modules of the MBF supports regulating the flow of traffic controllers in aiding decision making. Through the development of this tool controllers can acquire the knowledge that will help them to make better decisions. The research also presents the results of a simulation with two policies: equal flow distribution and prioritized. As an example it is shown that the separation of the times of departures may be reduced from 30% to 60%, depending on the policy.



A model based on artificial intelligence for knowledge management applied to the process of software development

Developing software is a common activity for many companies in various areas of modern society, even if that is not the focus of their business. This reality is present by the realization that most products and services are supported by computing systems, especially when talking about large companies and processes that involve any significance values​​. In the financial sector, for example, have organizations that devote a few thousand professionals to develop systems to automate their activities and the maintenance of others who have been employed for a long time.
In this context, the process of software development have been the subject of constant concern by the government, but most of the work is back in control of resources and the improvement and standardization of current processes, neglecting much of the knowledge that is present in this cycle. This research focuses attention on this issue, proposing a model based on artificial intelligence formalisms, which seeks to add mechanisms for the management of knowledge involved in the process of software development. The model described here has been constructed and applied to a case study in a large financial institution, obtaining promising results regarding its use as a way to increase the reuse of solutions already developed, avoid duplication of efforts in building similar solutions and also provide an effective alternative for the traceability of features and details of such solutions.



The solution was modeled as a Decision Support System (Decision Support System - DSS), through the construction of Submodule Modeling and Projection of Impact (MPI), part of Module Evaluation and Decision Support (BHAG), which comprises the modular System integration and Application Management measures to Control Air Traffic Flow (SISCONFLUX). The architecture of this module uses an autonomous agent that, in situations of congestion, provides suggestions for measures restrictive to air traffic controllers, focused on the problem of waiting on the ground (Ground Holding Problem - GHP). The agent acquires knowledge by the environment, through the algorithm QLearning, assessing the situation of the air resulting scenario restrictive measures suggested and restrictive measures. The evaluation of scenarios resulting air is carried through a function that computes data such as congestion level, time delay imposed on the aircraft, and an index of financial impact and equity in the distribution of the restrictive measures.

The results using data from a real environment, formed by setting the air Flight Information Region of Brasilia (FIR-BS), managed by First Integrated Center for Air Defense and Air Traffic Control (CINDACTA I). Analysis of these results indicates that factors such as equity and financial costs can be used in conjunction with data congestion, without hurting safety standards, where the agent learns to suggest actions to the human controller, which take into account the impact generated by the actions taken .



This paper proposes as a solution to this problem, a model based on graph theory that employs multiple algorithms for maximum network flow. The model was created from a software prototype called Modulo Balancing Flow (MBF). The modeling of the entire solution is Brazilian air space through a network flow. The model performs the balancing traffic flow and distribution of aircraft by means of maximum flow algorithms Edmonds-Karp, Dinic, FIFO Preflow Highest Label Preflow Push Push.

The validation of the implementations of algorithms is performed based on tests using actual data from air movements to compute the balancing flow. The model uses two distribution policies slack in controlled spaces. The first policy distributes the slack equally among sectors of airspace, the second policy distributes the slack in order to prioritize specific routes.

The tests employ two databases with days of aerial movements of high and low flow of traffic in the years 2008 and 2009. The data for the formation of the real scenario of Brazilian air space were provided by the Air Navigation Management Center (CGNA). The analysis of the test results indicate that the waiting time on the ground the aircraft can be reduced in most cases without causing congestion in the sectors of airspace. In addition, the use of multiple algorithms proposed different options load distribution of traffic flow.


18. ANTONIO CRESPO, 26/10/2010

Under the system of the Brazilian Airspace Control (SISCEAB), the activity Management Air Traffic Flow (ATFM) is of crucial importance, especially when considered two extremely important aspects, namely: first, the impact of ATFM (tactical) activity in air traffic control, including the inherent security implications of operations, and, second, the possible consequences of ATFM measures on the airport logistics. Therefore, it becomes imperative to establish processes and / or systems development (computational agents) that help traffic flow managers (human agents) to take action optimized. In this context, this paper presents a comparative analysis of the control of air traffic flow generated by a computational agent (intelligent) based on reinforcement learning. The objective of this agent is to establish delays in schedules takeoff of aircraft departing from TMA determined, so that the sectors of air traffic control not congest or saturate, due to an imbalance and eventual momentary between demand and capacity. The paper includes a case study which compares the measurements generated by the agent autonomously generated and measures taking into account the experience of managers crowded air traffic flow in the Management Center of Air Navigation (CGNA).



Uma proposta de modelagem ontológica para a Nomenclatura Comum do Mercosul (NCM). 15/08/2011. Dissertação (Mestrado em Sistemas Mecatrônicos) - Universidade de Brasília, . Orientador: Li Weigang.


A Nomenclatura Comum do Mercosul (NCM) é uma taxonomia utilizada pelo Brasil para a classificação de produtos, classificação esta considerada de grande importância no processo de importação e exportação de mercadorias, assim como nas operações de mercado interno. No entanto a NCM, da forma que está, não permite que a classificação de mercadorias seja uma prática simplificada, pois a atual nomenclatura utilizada no Mercosul encontra-se com os nome dos produtos desatualizados, os quais acabam por gerar sentidos dúbios. Uma forma de solucionar o problema seria com utilização do conceito de web semântica, que propicia a coleta de dados automáticos pela internet, e a ontologia, instrumento capaz de conceber novos sentidos aos termos existentes na NCM. Assim, a proposta de ontologia deste trabalho tem por objetivo fornecer novos domínios de interesse aos produtos da NCM, o que inclui sinônimos, línguas adicionais, restrições de comercialização dos produtos (importação e/ou exportação) bem como a possibilidade de correlacionar as leis existentes com os respectivos produtos. O estudo de caso realizado comprovou a eficiência da ontologia, uma vez que forneceu respostas esperadas a todos os questionamentos até então pendentes. Desta forma, comparando a atual taxonomia utilizada pelo Mercosul na classificação de seus produtos com a proposta deste trabalho, pode-se afirmar que a utilização do OntoNCM otimizou a forma de uso da Nomenclatura Comum do Mercosul.


20. ZHENG JIANYA, 09/03/2012

PageRank e W-entropia para analisar a influencia dos membros das redes sociais. 09/03/2012. Dissertação (Mestrado em Informática) - Universidade de Brasília, Coordenação de Aperfeiçoamento de Pessoal de Nível Superior. Orientador: Li Weigang.
With the fast development of the social networks, a suitable method should be developed to determine the influence of the people or brands in the communication of the modern e-society. Every social network has its ranking manner to display who is the most popular in that virtual society. However, this measurement is still incomplete and one-dimensional, such as the list ranked by the number of visits. In this research a new measurement model, W-entropy, is developed using information theory to study the influence of the individuals on multi-social networks. The real data are collected from Facebook, Twitter, YouTube and Google of the Internet. For comparing the effectiveness of the developed method, Famecount ranking is studied together using the same data. As the result, W-entropy method is suitable to reflect the uneven information distribution across the multi-platforms. This method is implemented in web ( to automatically present the ranking list of the related social network stars.



Modelagem de apoio a decisão para o problema de espera no ar utilizando sistemas multiagentes e aprendizagem por reforços. 08/03/2012. Dissertação (Mestrado em Informática) - Universidade de Brasília, . Orientador: Li Weigang.

O Módulo do Problema de Espera no Ar (AHPM) é proposto como um Sistema de Suporte à Decisão para auxiliar os controladores de tráfego aéreo em seu gerenciamento. Este sistema é desenvolvido utilizando uma plataforma Multiagente para organizar e aperfeiçoar as soluções para os controladores lidarem com o fluxo de tráfego no espaço aéreo brasileiro. O artigo foca na resolução de problemas usando Aprendizagem por Reforço para mostrar como a experiência histórica dos controladores pode ser usada para auxiliar em novas situações. Foram estudados cerca de 1.110 voos diários sob a responsabilidade de duas Regiões de Informação de Voo (FIR), FIR-Brasília e FIR-Recife considerando as funções de avaliações locais e globais e três intervalos de tempo para aplicar medidas restritivas. Como resultado, com a utilização do AHPM, o processo de decisão obteve melhoria com um elevado nível de automação. Ao mesmo tempo, o congestionamento nos setores envolvidos foi reduzido de 15% à 57% em cenários locais e de 41% à 48% no cenário global.