Abstract

Abstract

Abstract
This paper proposes a reinforcement learning model for intelligent energy management in buildings, using a UCB1 based approach. Energy management in buildings has become a critical task in recent years, due to the incentives to the increase of energy efficiency and renewable energy sources penetration. Managing the energy consumption, generation and storage in this domain, becomes, however, an arduous task, due to the large uncertainty of the different resources, adjacent to the dynamic characteristics of this environment. In this scope, reinforcement learning is a promising solution to provide adaptiveness to the energy management methods, by learning with the on-going changes in the environment. The model proposed in this paper aims at supporting decisions on the best actions to take in each moment, regarding buildings energy management. A UCB1 based algorithm is applied, and the results are compared to those of an EXP3 approach and a simple reinforcement learning algorithm. Results show that the proposed approach is able to achieve a higher quality of results, by reaching a higher rate of successful actions identification, when compared to the other considered reference approaches.

Abstract

Abstract
The increase in retention of customer in gyms and health clubs is nowadays a challenge that requires concrete and personalized actions. Traditional data mining studies focused essentially on predictive analytics, neglecting the business domain. This work presents an actionable knowledge discovery system which uses the following pipeline (data collection, predictive model, loyalty actions). In the first step, it extracts and transforms existing real data from databases of the sports facilities. In a second step, predictive models are applied to identify user profiles more susceptible to dropout. Actionable rules are generated based on actionable attributes that should be avoided, in order to increase retention. Finally, in the third step, based on the previous actionable knowledge, experimental planning is carried out, with test and control groups, in order to find the best loyalty actions for customer retention. This document presents a simulation and the measure of the business utility of an actions sequence to avoid dropout. © 2019, Springer Nature Switzerland AG.

Abstract
Conducting scientific experiments in driving simulators requires the modeling of reliable and complete road environments. These environments must provide extensive landscapes with the artifacts and natural element that can be usually found in the real world. This paper presents a method to efficiently produce models of natural vegetation. The produced models are then applied to populate existing terrain definitions, allowing the fast preparation of extensive environments with realistic landscapes. The human supervisor can interact in this generation process, in order to obtain custom landscapes definitions. After the landscape generation process, the road network definition can be then generated, producing a complete driving environment, in an integrated modeling process. The proposed method allows modeling a wide range of drive environments, with the realism and quality required to the realization of virtual training or experimental work in many terrain based activities, such driving simulators.