M.Sc. Thesis


Minea Filipec, Genetic Algorithm for Optimal Design of Link Structured Distribution Networks, M.Sc. Thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Departement of Power Systems, Zagreb, Croatia, September 1999.



In M.Sc. thesis a new method based on genetic algorithms is presented for optimal design of spatially closed distribution networks in urban areas. With regard of distribution network planning procedure, optimal design of distribution networks is one of the most important steps due to the influence it has on decisions regarding optimal locations and sizing of new HV/MV substations, sizing of new transformers which replace transformers or are added in the existing HV/MV substations, and network routing. Genetic operators and chromosome representation which are used in genetic algorithm enable optimal distribution network design in predefined structure in observed area as a whole with respect to the investment and operational costs. Existing network, voltage drop calculation (both in normal and outage conditions), power and energy loss calculations, and cable and transformer thermal limit checking are included in the optimization procedure. The method is based on variant solutions proposed by computer and additionally evaluated by planer. The method can be applied either to short or long term planning strategies.

The thesis is divided into the eleven chapters.

In the introductory chapter classification of characteristics of distribution network design problem is stated. Also the role and importance of planning of medium voltage distribution networks in urban area is explained.

The second chapter begins with classification and historical review of well-known optimization techniques. Then, the advantages of genetic algorithms are explained that they have in comparison to other optimization algorithms, especially simulated annealing and hill climbing algorithms. After the description of features of medium voltage distribution network planning problem, the review of methods developed for distribution network planning problem solving is presented. The classification of methods in consideration of network structures to which they are applied, planning period, existing network elements taken into account, and forecasted peak load is separately presented. Next, optimization methods are classified based on algorithms they use. The chapter ends with discussion of goal, requirements and considerations characteristic for the methods developed for planning of spatially closed distribution networks, which are subject of this thesis.

The third chapter deals with physical properties of distribution networks. Following the description of structures of customary distribution networks, with emphasis on structures modeled in this thesis, operational features of these structures are considered in detail. Next, requirements to be satisfied like voltage drop, maximal load and number of MV/LV substations per distribution network link, cable and transformer thermal limit are discussed. Then, methodology for economic analysis of annual operational and investment costs and the evaluation of possible solutions based on actual value method is presented. Also general considerations relevant to the optimization procedure are mentioned. The review of methods is given for peak load forecasting which results are adequate for planning procedure considering the model of distribution network used. Existing network element parameters, necessary for distribution network design algorithm, are presented. In particular, information systems are examined, as well as respective models of geographic area and existing distribution network elements which more or less accurately comprise urban properties, but all can be used as database of necessary data about existing network elements. The chapter ends with description of an auxiliary algorithm of the data preparation procedure which enables the selection of number of links between pairs of HV/MV substations taking into account thermal capacity and number of available feeders in the HV/MV substations.

At the beginning of fourth chapter the review of basic components and terms of natural selection is given. After short description of classical genetic algorithm’s steps, operators and chromosome representation, the mathematical theory of convergence of classical genetic algorithm to global optimum is briefly examined. Then, the review of evolution algorithms is given which simulate the natural evolution and genetic, as much as classical genetic algorithms, and are developed for specific optimization problems. Special part of fourth chapter is devoted to description of operators and parameters of genetic algorithm for optimal design of distribution networks presented in this thesis. Chromosome representation and procedure of chromosome decoding are emphasized which enable design of distribution networks with multiple HV/MV substations in predefined structure without unnatural decomposition of the problem into number of subproblems. Besides, the description of population initialization, chromosome fitness evaluation, and selection procedures are presented. The influence of different penalty functions, crossover and mutation operators and genetic algorithm parameters (population size, selection intensity, mutation probability, number of evolution cycles) on genetic algorithm convergence is analyzed.

In the fifth chapter the comparison of developed genetic algorithm with other available algorithms is carried out based on the results of simple link distribution network design problem. After short description of flow of optimization procedure when applied on real distribution network planning, the algorithm’s applicability and efficiency is examined in two examples of complex distribution network design problem of central area of city of Zagreb.

Other chapters deal with conclusions, summary, and list of symbols and references. 


power system, primary distribution. link structured network, optimization, genetic algorithm