Economic dispatch in isolated networks with renewables using evolutionary programming L. M. Proença (2,3) J. Luís Pinto (2) Manuel A. Matos (1,2) lproenca@inescn.pt jpinto@duque.inescn.pt mmatos@inescn.pt (1) FEUP - Faculdade de Engenharia da Universidade do Porto, Porto - Portugal (3) ISLA – Instituto Superior de Línguas e Administração - Vila Nova de Gaia - Portugal (2) INESC, Instituto de Engenharia de Sistemas e Computadores, Porto, Portugal Abstract - The paper presents a new technique for the Economic Dispatch of isolated power systems with a high penetration of renewable forms of energy. It also describes a software module based on this technique integrated on a real- time control system. Keywords: Economic Dispatch, Evolutionary Programming, Renewable Energy, Isolated Systems. I. INTRODUCTION The paper addresses the problem of dispatching a medium- size isolated power system where the penetration of renewable sources, especially wind power, is considerable. It is the case of many islands that have favorable conditions to use wind power combined to a conventional generating system with Diesel units and gas turbines, and sometimes steam turbines. This problem is similar to the conventional economic dispatch problem, but most cost functions of typical generating units are not well-behaved (non-convex functions), and there may also exist the need to satisfy additional constraints on security and spinning reserve that are generally difficult to include in analytical formulations. In these circumstances, the use of Evolutionary Programming [1,5-7], a variety of Evolutionary Algorithms [2,3], that relies on mutation rather than crossover [4], is an attractive hypothesis, due to the inherent flexibility of using a fitness function (to evaluate each candidate solution) that is not constrained regarding convexity, etc. The ease of codification and implementation of the method and its implicit parallelism lead to fast prototyping and good performances regarding execution times. The paper reports the development of an Economic Dispatch module in the framework of the project CARE - "Advanced control advice for power systems with large scale integration of renewable energy sources" (ENN- JOULE program). Partners are NTUA (Greece), INESC (Portugal), ARMINES (France) and DEI (Greece), the Public Power Corporation of Greece, and tests are made with the Power System of Crete. The module follows the usual steps in an Evolutionary Programming algorithm, where the solutions are not coded in chromosomes (as in Genetic Algorithms). The algorithm operates directly over the solutions by small mutations, but also uses a selection process based on elitism. Paper BPT99-361-25 accepted for presentation at the IEEE Power Tech’99 Conference, Budapest, Hungary, Aug 29-Sept 2, 1999. Solutions are defined as generation values for all the units in the system and initially are identical to the pre-dispatch values proposed by an unit commitment algorithm run previously. The fitness function assesses the quality of the proposed solution setting the basis for the selection process. It proposes a penalty for constraint violations (voltage levels in PQ buses and line load limits) that increases with the number of generations, in order to allow initially the digression of the algorithm through unfeasible parts of the universe of possible solutions. It also contains processes of auto-correction of power losses and voltage levels. Furthermore, a process to evaluate the dynamic security of a proposed solution can easily be added to the global cost calculation, provided that a fast tool for security assessment is available (neural networks and decision trees are successfully used for that in CARE). This possibility is especially important when wind penetration is strong and weather conditions are bad, due to the isolated nature of the networks. II. EVOLUTIONARY PROGRAMMING Evolutionary Algorithms (EA) are computer-based problem-solving systems based on principles of evolution theory. A variety of EA have been developed and they all share a common conceptual base of simulating the evolution of individual structures via processes of Selection, Mutation and Recombination. The processes depend on the perceived performance of the individual structures as defined by an environment. The interest in these algorithms has been rising fast for they provide robust and powerful adaptive search mechanisms. The interesting biological concepts on which EA are based also contribute to their attractiveness. There has been a great interest in the use of EA in Power Systems [8] because these approaches are very well suited to deal with all those kinds of problems that usually represent nightmares for researchers and developers: integer variables, non convex functions, non differentiable functions, domains not connected, badly-behaved functions, multiple local optima, multiple objectives, etc. Furthermore, they are not necessarily restricted to deal with numerical models, allowing the natural building of hybrid models including knowledge, under the forms of rules or other. This complexity is what is required, in order to build larger Power System models with more adherence to reality. In very complex situations, they seem to be the only