A Technical and Distributed Management Basis for an Environmentally Clean and Sustainable Energy Supply

Abstract

Rising market prices for energy, an apparent future shortage in fossil fuels, and alarming reports on pollution through CO2 are causing a world-wide trend towards renewable and ecologically clean forms of energy. We report about ongoing work in the R&D project DEZENT establishing renewable electric energy supply and eventually replacing fossil energy sources. Producers are at the same time also consumers. Their production and consumption are largely unpredictable. With our combined expertise in Real-Time systems and Electric Power Distribution we developed price negotiations which are pursued by consumer/ producer agents on a P2P basis and are governed by tough end-to-end deadlines (< 0.5 sec) dictated by EE constraints. The strategies used for periods of 0.5 sec are designed for fast convergence while we may at the same time assume a constant demand/ supply situation. Malicious users will not succeed, and customers pay considerable less than under conventional management policies or structures. In this paper we allow the negotiation strategies themselves to be adaptive across periods thus achieving a most flexible bargaining for each individual customer involved. For this purpose we have defined distributed learning algorithms derived from Reinforcement Learning. While maintaining all benefits from the earlier stage of development we demonstrate that we obtain a much better performance across periods than the initial static algorithms. To our knowledge we have presented and investigated the first distributed learning algorithm in the area of Adaptive Real-time Systems. Since the electric distribution management can be equally finalized within each period we have laid the ground for a thorough provision with sustainable and clean electric energy.