Suchergebnisse

The solution for a Multi-Objetive Reinforcement Learning problem is a set of Pareto optimal policies. MPQ-learning is a recent algorithm that approximates the whole set of all Pareto-optimal deterministic policies by directly generalizing Q-learning to the multiobjective setting. In this paper we present a modification of MPQ-learning that avoids useless cyclical policies and thus improves the number of training steps required for convergence. ; Supported by: the Spanish Government, Agencia Estatal de Investigaci´on (AEI) and European Union, Fondo Europeo de Desarrollo Regional (FEDER), grant TIN2016-80774-R (AEI/FEDER, UE); and Plan Propio de Investigación de la Universidad de Málaga - Campus de Excelencia Internacional Andalucía Tech.

AbstractPrices are increasingly set by algorithms. One concern is that intelligent algorithms may learn to collude on higher prices even in the absence of the kind of coordination necessary to establish an antitrust infringement. However, exactly how this may happen is an open question. I show how in simulated sequential competition, competing reinforcement learning algorithms can indeed learn to converge to collusive equilibria when the set of discrete prices is limited. When this set increases, the algorithm considered increasingly converges to supra‐competitive asymmetric cycles. I show that results are robust to various extensions and discuss practical limitations and policy implications.

The electricity markets restructuring process encouraged the use of computational tools in order to allow the study of different market mechanisms and the relationships between the participating entities. Automated negotiation plays a crucial role in the decision support for energy transactions due to the constant need for players to engage in bilateral negotiations. This paper proposes a methodology to estimate bilateral contract prices, which is essential to support market players in their decisions, enabling adequate risk management of the negotiation process. The proposed approach uses an adaptation of the Q-Learning reinforcement learning algorithm to choose the best from a set of possible contract prices forecasts that are determined using several methods, such as artificial neural networks (ANN), support vector machines (SVM), among others. The learning process assesses the probability of success of each forecasting method, by comparing the expected negotiation price with the historic data contracts of competitor players. The negotiation scenario identified as the most probable scenario that the player will face during the negotiation process is the one that presents the higher expected utility value. This approach allows the supported player to be prepared for the negotiation scenario that is the most likely to represent a reliable approximation of the actual negotiation environment. ; This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 703689 (project ADAPT) and No 641794 (project DREAM-GO); NetEfficity Project (P2020 − 18015); and UID/EEA/00760/2013 funded by FEDER Funds through COMPETE pro-gram and by National Funds through FCT. ; info:eu-repo/semantics/publishedVersion

Electricity markets are complex environments, which have been suffering continuous transformations due to the increase of renewable based generation and the introduction of new players in the system. In this context, players are forced to re-think their behavior and learn how to act in this dynamic environment in order to get as much benefit as possible from market negotiations. This paper introduces a new learning model to enable players identifying the expected prices of future bilateral agreements, as a way to improve the decision-making process in deciding the opponent players to approach for actual negotiations. The proposed model introduces a con-textual dimension in the well-known Q-Learning algorithm, and includes a simulated annealing process to accelerate the convergence process. The proposed model is integrated in a multi-agent decision support system for electricity market players negotiations, enabling the experimentation of results using real data from the Iberian electricity market. ; This work has received funding from the European Union's Horizon 2020 research and innovation programme under project DOMINOES (grant agreement No 771066) and from FEDER Funds through COMPETE program and from National Funds through FCT under the project UID/EEA/00760/2019.

Electricity markets are complex environments, which have been suffering continuous transformations due to the increase of renewable based generation and the introduction of new players in the system. In this context, players are forced to re-think their behavior and learn how to act in this dynamic environment in order to get as much benefit as possible from market negotiations. This paper introduces a new learning model to enable players identifying the expected prices of future bilateral agreements, as a way to improve the decision-making process in deciding the opponent players to approach for actual negotiations. The proposed model introduces a con-textual dimension in the well-known Q-Learning algorithm, and includes a simulated annealing process to accelerate the convergence process. The proposed model is integrated in a multi-agent decision support system for electricity market players negotiations, enabling the experimentation of results using real data from the Iberian electricity market. ; This work has received funding from the European Union's Horizon 2020 research and innovation programme under project DOMINOES (grant agreement No 771066) and from FEDER Funds through COMPETE program and from National Funds through FCT under the project UID/EEA/00760/2019. ; info:eu-repo/semantics/publishedVersion

Uncrewed autonomous vehicles (UAVs) have made significant contributions to reconnaissance and surveillance missions in past US military campaigns. As the prevalence of UAVs increases, there has also been improvements in counter-UAV technology that makes it difficult for them to successfully obtain valuable intelligence within an area of interest. Hence, it has become important that modern UAVs can accomplish their missions while maximizing their chances of survival. In this work, we specifically study the problem of identifying a short path from a designated start to a goal, while collecting all rewards and avoiding adversaries that move randomly on the grid. We also provide a possible application of the framework in a military setting, that of autonomous casualty evacuation. We present a comparison of three methods to solve this problem: namely we implement a Deep Q-Learning model, an ε-greedy tabular Q-Learning model, and an online optimization framework. Our computational experiments, designed using simple grid-world environments with random adversaries showcase how these approaches work and compare them in terms of performance, accuracy, and computational time. ; R.Y. is partially supported by NSF DMS 1916037 and Consortium for Robotics and Unmanned Systems Education and Research (CRUSER). ; NSF DMS 1916037 ; Consortium for Robotics and Unmanned Systems Education and Research (CRUSER)

Underwater Wireless Sensor Networks (UWSNs) have aroused increasing interest of many researchers in industry, military, commerce and academe recently. Due to the harsh underwater environment, energy efficiency is a significant theme should be considered for routing in UWSNs. Underwater positioning is also a particularly tricky task since the high attenuation of radio-frequency signals in UWSNs. In this paper, we propose an energy-efficient depth-based opportunistic routing algorithm with Q-learning (EDORQ) for UWSNs to guarantee the energy-saving and reliable data transmission. It combines the respective advantages of Q-learning technique and opportunistic routing (OR) algorithm without the full-dimensional location information to improve the network performance in terms of energy consumption, average network overhead and packet delivery ratio. In EDORQ, the void detection factor, residual energy and depth information of candidate nodes are jointly considered when defining the Q-value function, which contributes to proactively detecting void nodes in advance, meanwhile, reducing energy consumption. In addition, a simple and scalable void node recovery mode is proposed for the selection of candidate set so as to rescue packets that are stuck in void nodes unfortunately. Furthermore, we design a novel method to set the holding time for the schedule of packet forwarding base on Q-value so as to alleviate the packet collision and redundant transmission. We conduct extensive simulations to evaluate the performance of our proposed algorithm and compare it with other three routing algorithms on Aqua-sim platform (NS2). The results show that the proposed algorithm significantly improve the performance in terms of energy efficiency, packet delivery ratio and average network overhead without sacrificing too much average packet delay.