Search

A novel structure of matrix-converter-based Smart Transformer (MC-ST) is proposed for hybrid microgrid applications. The conventional ST structure in hybrid microgrid normally consists of two different kinds of modules: cascaded H-bridges (CHBs) and dual active bridges (DABs). For reliability and simplicity, the modularity concept is employed in the proposed structure to obtain only one kind of module. Utilizing the feature of matrix converters as the universal converters, the proposed ST can transfer AC input to AC output as well as to DC output using the same module and same modulation method. Moreover, in order to implement the proposed ST, feedback control of low voltage AC (LVAC) and low voltage DC (LVDC) is implemented. Simulation results show that the proposed MC-ST is feasible and has a potential application in the future hybrid microgrid.

Electronic health records (EHRs) contain patient-related information formed by structured and unstructured data, a valuable data source for Natural Language Processing (NLP) in the healthcare domain. The contextual word embeddings and Transformer-based models have proved their potential, reaching state-of-the-art for various NLP tasks. Although the performance for downstream NLP tasks with free-texts written in English has recently improved, less resource is available considering clinical texts and low-resource languages such as Portuguese. Our objective is to develop a Generative Pre-trained Transformer 2 (GPT-2) language model for Portuguese to support clinical and biomedical NLP tasks. We fine-tuned a generic Portuguese GPT-2 model to corpora of biomedical texts written in Portuguese, using transfer learning. We experimented on a public dataset, manually annotated for detecting patient fall, i.e., a classification task. Our in-domain GPT-2 model outperformed the generic Portuguese GPT-2 model by 3.43 in F1-score (weighted). Our preliminary results show that transfer learning with domain literature can benefit Portuguese biomedical NLP tasks, aligned with other languages' results.

This paper presents a novel active voltage regulation transformer (AVRT), to guarantee on-load voltage regulation of AC microgrids, even when supplying unbalanced loads. The proposed AVRT includes a 4-leg AC/AC matrix converter and a series transformer to allow the voltage regulation in the low voltage (LV) AC microgrid. Using the space vector representation, a sliding mode control approach is proposed for the 4-leg matrix converter, and the voltage in the microgrid is then controlled using a linear control approach. Obtained results show that the proposed system regulates the voltage in the LV microgrid even when the loads are unbalanced. Additionally, it allows regulation of the voltage in the microgrid in case of a voltage sag or an overvoltage in the medium voltage (MV) side.

This paper uses non-traditional data, from a MicroFinance Institution (MFI), in a Credit Scoring loan classification problem and addresses a common problem in emerging markets of the lack of a verifiable customers’ credit history. We perform a set of experiments to define a baseline model and prove the relevance of node embedding features, in credit scoring models, using a real world dataset.

This paper presents a novel control approach for a Dual Active Bridge (DAB) converter interconnecting a telecom isolated DC microgrid (IDCM) to a 380 V DC grid. A DAB converter is the ideal solution for this interconnection because it provides galvanic isolation, allows reversible power flow, and can operate with inherent soft switching, leading to high efficiency performance. The novel controller is based on Lyapunov's stability theory and adds integral terms to stabilize the IDCM nominal voltage. The new controller ensures the IDCM fast and stable transient response upon the connection of constant power loads (CPLs). In order to highlight the advantages of the new controller, comparisons using MATLAB/Simulink simulations are made with cascaded voltage/current linear controllers, their limitations being shown. Results point out that the novel controller is able to deal with the negative incremental resistance of CPLs, without the need of bulky stabilizing DC capacitors.

Organizations have been increasingly relying on software development to manage their business, making it an essential activity. Companies have been using collaborative development environments to speed up their deliveries, resulting in projects tending to be produced by developers from different parts of the world with different cultures. In this context, controlling the software's overall quality by ensuring code quality is a crucial aspect of software development. Software quality can be analyzed based on software parameters like complexity, reusability, testability and maintainability. Such parameters are associated with different group of code metrics which are sets of software measures, providing better insights to developers about their code. In this research we aim to support software developers in better understanding code metrics information. We present a visual tool based on an original method which applies code metrics to an open source collaborative software development environment, grouping them into software product quality parameters. Developers are provided with dashboards and recommendations related to the part of code requiring quality improvement.

Despite the great number of Systems of Systems (SoS) being developed, building them still remains hard and difficult. Currently, there is a lack of methods capable of supporting architects for building an actual SoS. In this paper we introduce an original method called GAMBAD for developing an SoS from a practical point of view. Our method guides the development of SoS on top of a multi-agent layer supported by ontologies. We tested GAMBAD by building an SoS in the domain of Health Care. Early results show that by using our method, architects can develop an SoS faster and more accurately.

In this paper, we present a Machine Learning approach based on commitment to deal with two risky situations over game usage lifecycle: the prediction of churn and players remaining lifetime. These risky situations are gauged by game producers to try to maintain the players motivated, intervening when players tend to leave. The problem is that this information is not trivial to obtain due to the players‘ motivational usage linked to the many activities available in the game content (e.g., all the possible actions in an MMORPG). To deal with that, we proposed the use of the commitment concept to assign the engagement of a player to a game based on the usage data. The main aspect of the proposed approach is related to the preprocessing step, where measures for the actions related to commitment were computed, generating a tendency degree for each one. We used a dataset from the CIG 2017 Game Data Mining competition which focused on the same risky situations as a comparison baseline. This paper approach overcomes the competition’s results for both challenges. All experiments followed the competition rules. These experiments show that looking at tendencies on commitment is an option to gauge players’ engagement over time.

Although the concept of System of Systems (SoS) has become quite popular, most applications are still hand crafted. In this paper we present a framework, called MBA for Memory-Broker-Agent, addressing the development of systems of systems from an engineering perspective. The main features of the framework result from the experience gained from building an SoS for developing software collaboratively. In the paper we recall the requirements for building an SoS and show how they can be met by using a multi-agent substrate. The MBA framework is a first step towards proposing a generic platform for developing systems of systems.

The information and the way we use it are the key for an efficient implementation of a management procedure. It is imperative that information is delivered promptly, in a layout accessible to everyone. This way, the concept Indicators of Performance and Dashboards has been created. It concerns a visual imperative technique for information analysis and decision support. Ten interviews were conducted with managers and employees of accounting and auditing companies to understand the state of knowledge and use of the concepts. Thus, eight indicators have emerged that complement the strategic dashboard, addressed to management. This study contributes to expand the managers' knowledge in accounting and audit sectors about development evaluation through dynamic tools like the dashboard. It allows the management to access a daily life support tool and the monitoring and realization of their goals.

People involved in software development seek better ways of developing quality software. As software development is a collaborative activity dependent on technology and performed by group of people, the quality can be directly linked to the degree of collaboration and commitment of the development team. This research aims to identify and analyze software quality metrics that can measure software development team members' participation regarding their contribution in the project source code. Analyzing such information can contribute with software engineering project managers to conduct and better organize project development teams. To achieve this goal, we defined a set of quality metrics, applied them to open source code and analyzed them in other to evaluate members' contribution. Results show that it is possible to determine whether the team member contribution increase, decrease or had no influence on source code quality.

This paper presents the design, control and analysis of a small DC residential grid supplying constant power loads. A DC/DC Buck converter is used to connect the main DC bus to the DC distribution grid. Constant power loads are DC/DC converters embedded in equipment, connected to the DC distribution grid, and considered to be operated at constant power, as usual in most electronic equipment. Two possible solutions to control the main DC/DC converter are designed and assessed: a linear controller and a novel non-linear controller based on the Lyapunov direct control method. The two control strategies are then compared, considering response time, overvoltage, and overall disturbance in the DC grid voltage. The obtained results show a better performance of the Lyapunov non-linear controller.

International audience; Collaborative software development is a complex activity. An important factor that needs to receive attention in collaborative software development is software quality. High quality software reduces the development and the maintenance; improves delivery schedules; and reduces repairs and rework. In order to measure, evaluate, control and improve the software quality, software metrics can be used. In this research we present an advanced collaborative environment for software development currently being built, called ACE4SD, which intends to support the improvement of the code quality during collaborative software development. ACE4SD is a system of systems composed of a software development environment, a multi-agent system and a platform to capitalize and manage knowledge, all of them being integrated in the same environment. ACE4SD can provide personalized support to team members to improve the code quality and encourage its reuse, it can answer questions or doubts arisen during the development, record document problems and solutions, and improve the awareness and collaboration between the participants.

Inspired on decision trees and evolutionary algorithms, this paper proposes a learning algorithm of constructive neural networks that relies on three principles: to layout the neurons in a tree-like structure; to train each neuron individually; and, to optimize all the weights using an evolutionary approach. This way, it is expected to advance in two main questions concerning multilayer perceptrons (MLPs): how to determine the network architecture and how to build models that are more comprehensible. Based on the normalized information gain of each attribute, the algorithm builds the network architecture. In the process, it automatically creates a set of training examples for each individual neuron and executes single-cell learning. Once the network is created and trained, particle swarm optimization is utilized to evolve the connections of the network. Five metrics were utilized to validate the method when compared to decision trees and MLPs: accuracy, sensitivity, specificity, precision and comprehensibility. The experiments were executed in thirteen different databases and the results suggest that the proposed algorithm can generate neural networks with good classification performance and more comprehensible.

International audience; With the emergence of Docker, it becomes easier to encapsulate applications and their dependencies into lightweight Linux containers and make them available to the world by deploying them in the cloud. Compared to hypervisor-based virtualization approaches, the use of containers provides faster start-ups times and reduces the consumption of computer resources. However, Docker lacks of deployability verification tool for containers at design time. Currently, the only way to be sure that the designed containers will execute well is to test them in a running system. If errors occur, a correction is made but this operation can be repeated several times before the deployment becomes operational. Docker does not provide a solution to increase or decrease the size of container resources in demand. Besides the deployment of containers, Docker lacks of synchronization between the designed containers and those deployed. Moreover, container management with Docker is done at low level, and therefore requires users to focus on low level system issues. In this paper we focus on these issues related to the management of Docker containers. In particular, we propose an approach for modeling Docker containers. We provide tooling to ensure the deployability and the management of Docker containers. We illustrate our proposal using an event processing application and show how our solution provides a significantly better compromise between performance and development costs than the basic Docker container solution.

This paper introduces the use of a modified feedforward neural network to cope with the problem of predicting protein functions. Since this kind of classification task is inherently hierarchical, this work proposes the use of two different architectures for the modified feedforward neural network, both mimicking the hierarchical nature of the classes (protein functions) to be predicted. The first approach consists of four feed-forward neural networks in cascade, each one taking as input the classification obtained by the previous network, which means, the input to a network is the classes that could be assigned to the protein at the immediately higher (parent) level in the class hierarchy. The second approach is an extension of the first one, which also adds as input to each sub-network the attributes of the protein being classified. In both situations, it was used two kinds of feed-forward architectures: an Adaline network, which is composed of a single layer of adjustable weights, and a MLP ("Multi-Layer Perceptron"), composed by two layers of adjustable weights. Both approaches were compared with a baseline consisting of a single MLP that maps the input attributes to the classes of the lowest level in the hierarchy. The MLP was built with the input layer, plus one hidden layer and one output layer. The three approaches were compared on eight datasets, the first four involving the prediction of GPCR (G-Protein Coupled Receptor) functions and the second four datasets involving the prediction of enzymes functions. The results show that a big-bang hierarchical neural network, based on the MLP paradigm, using a top-down evaluation for new instances has better behavior in hierarchical problems, when compared to its flat version.

This article presents a method to assess whether folksonomies are useful shared conceptualizations for conceptual modeling. A method based on the tripartite model (actors, concepts and instances) induces from social applications a folksonomy related to a domain of knowledge. The hypothesis of this paper is that if folksonomies are shared conceptualizations, then using them in conceptual modeling should reduce the number of divergences between actors when they elicit terms to be part of a model (a concept map in this article). A controlled experiment of conceptual modeling was performed with experimental groups who received tags inducted from data of categorizations from Delicious and control groups who received terms extracted from Web pages categorized on Delicious. The results show that the experimental groups obtained less divergences (on average) in elicitation of terms when compared to control groups.

This paper presents the FOLKUS-SD a module of CSCW-SD, intend to build Folksonomies from source codes. The CSCW-SD is an architecture to integrate tools in a development environment through a Multi-Agent System. The Folksonomies are built using dynamic data collected during codification. The Folksonomies' entities are represented by the module as: the users are the developers, the tags are names of classes, methods and attributes, and the resources are source codes. The Folksonomies can be used for software documentation. The Folksonomies produced by FOLKUS-SD can be used in the software documentation for monitoring the status of some project and to get quantitative data about the project and its participants.