Vous êtes sur la page 1sur 8

750

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 2, MAY 2005

A Multiobjective Evolutionary Algorithm for the Sizing and Siting of Distributed Generation
Gianni Celli, Member, IEEE, Emilio Ghiani, Susanna Mocci, Member, IEEE, and Fabrizio Pilo, Member, IEEE
AbstractIn the restructured electricity industry, the engineering aspects of planning need to be reformulated even though the goal to attain remains substantially the same, requiring various objectives to be simultaneously accomplished to achieve the optimality of the power system development and operation. In many cases, these objectives contradict each other and cannot be handled by conventional single optimization techniques. In this paper, a multiobjective formulation for the siting and sizing of DG resources into existing distribution networks is proposed. The methodology adopted permits the planner to decide the best compromise between cost of network upgrading, cost of power losses, cost of energy not supplied, and cost of energy required by the served customers. The implemented technique is based on a genetic algorithm and an -constrained method that allows obtaining a set of noninferior solutions. Application examples are presented to demonstrate the effectiveness of the proposed procedure. Index TermsDistributed generation, distribution networks, genetic algorithms, multiobjective programming.

I. INTRODUCTION

OWADAYS, the need for more exible electric systems, the changing in the regulatory and economic scenarios, the importance of harnessing energy savings and minimizing environmental impacts have been providing the impetus for the development of distributed generation (DG). For these reasons, DG is predicted to play an increasing role into the electric power system of the near future [1][6]. Whether DG is properly planned and operated it may provide benets to distribution networks (e.g., reduction of power losses and/or deferment of investments for network enforcing, etc.), otherwise it can cause degradation of power quality, reliability, and control of the power system [5]. Thus DG offers an alternative that planners should explore in their search for the best solution to electric supply problems and requires new planning paradigms and procedures able to face a more complex and uncertain scenario [4], [7], [8]. Ault et al. in [9] have pointed out the dichotomy between the advanced status of academic researches on planning and the unwillingness of companies to resort to such algorithms. Indeed, the planners need tools to deal with uncertainties, risks, and multiple criteria. The nal choice will be subjectively operated in the set of good solutions [9]. The idea behind the paper is to deal with both questions by taking into consideration the need of planning tools for the acManuscript received May 11, 2004; revised November 14, 2004. This work was partly supported by the Italian Ministry for Productive Activities (MAP) under a Public Interest Energy Research Project named Ricerca di Sistema (MAP Decree of February, 28th 2003). Paper no. TPWRS-00251-2004. The authors are with the University of Cagliari, 09123 Cagliari, Italy (e-mail: celli@diee.unica.it; pilo@diee.unica.it). Digital Object Identier 10.1109/TPWRS.2005.846219

commodation of DG in a distribution system according to the planners point of view. The starting point has been the observation that a planning tool should incorporate some of the multiple drivers in current electricity distribution planning practice (e.g., reliability aspects, DG integration, economic and technical aspects of competitive electricity markets, etc.). The second important point is that such a tool should leave the planner the faculty of choosing which aspects to consider in his search for the optimal solution. This is particularly important in the problem of the optimal siting and sizing of DG, where the distinctiveness of the planning objectives, which may conict each other (e.g., investments cost vs. the cost of the expected energy not supplied), has a tremendous impact on the search for a feasible conguration. Therefore, this optimization has been stated as a multiobjective (MO) planning problem. The MO permits a better simulation of the real world, often characterized by contrasting goals, and gives the planner the capability of making the nal decision by selecting, on the basis of his individual point of view, the most trade-off solution in a wide range of suitable solutions. The proposed MO optimization procedure makes use of a genetic algorithm (GA) proposed by the authors in [10][12] and it is constituted by a three steps process that can be described as follows. 1) Find a global noninferior solution for the MO problem with the GA. 2) Adopt the -constrained technique to create a set of noninferior solutions with an iterative procedure. 3) Choose the best compromise solution in a set of feasible solutions. The structure of the paper is the following. In Section II, the evolution of distribution system is delineated as well as the need of new planning and design tools. In Section III, IV, V, and VI the description, formulation and implementation of the methodology are proposed. In Section VII the results obtained by the application of the methodology are deeply discussed to highlight the usefulness of the MO approach. II. DISTRIBUTION NETWORK OF THE FUTURE Until now, distribution networks have been regarded as a passive termination of the transmission grid. Their task is to provide energy reliably and efciently to the nal users connected to the medium and low voltage networks. Distribution networks are congured according to the radial scheme, with unidirectional power ows and with simple protection equipment that permits the safe and economical operation of the power system. A greater penetration level of DG into existing systems will

0885-8950/$20.00 2005 IEEE

CELLI et al.: A MULTIOBJECTIVE EVOLUTIONARY ALGORITHM

751

change completely this well consolidated environment into a new one, where distribution networks will be no longer passive. Hence, a gradual but ineluctable changing toward a new kind of active networks has been foreseen, with DG units actively involved in system management and operation [13][19]. In active networks intentional islanding, the ability of managing an unfaulted portion of the network by using DG to enhance the service provided thereby, plays a key role. New control algorithms for rapid network reconguration, new communication protocols for data exchange between generators and loads, and reliable communications systems are essential to allow that such innovative practice can take place efciently [15][21]. Drivers and political decisions are essential to provoke and support such a drastic and revolutionary development, which will take many years to be completed. Anyway, the rst political directives have been recently promulgated to favor the changing [22]. Indeed, if DISCO is allowed to own its generation, DG may be a valuable option to defer investments and to improve the quality of service. If only private generators are allowed, DISCO may resort to planning methodologies to evaluate any additional credit that might be offered to private investors who decide to place generators in appropriate locations of the network. By so doing, DISCO may solve some specic problems (e.g., improving the voltage prole or reducing the frequency and the duration of sustained interruptions with the intentional islanding) minimizing the investments to upgrade the network [7], [8]. This is the Italian current situation, where some regulated DISCOs purchase energy from the wholesale liberalized energy market to feed their retail customers at a xed price all over the country. Regardless of the fact that Italian DISCOs are currently not allowed to own generators, DG, as well as other distributed resources, is a feasible option that should be considered and compared to traditional power supply. According to this need, the proposed tool aims at solving DG siting and sizing problems, allowing the planner to improve a particular term of the global network cost by taking into account uncertainties in power production [10], [11]. Furthermore, in a liberalized energy market the methodology can be used to assess the economically adopted network. Cost and benets of the real network can be allocated to loads and generators by comparing the effective situation to the reference network [22]. III. MULTIOBJECTIVE PROGRAMMING The new scenario of power system forces a change in duties and objectives of traditional planning and it compels to take into account several objectives that are in mutual conict. The use of MO methodologies gives information on the consequences of the decision with respect to all the objective functions dened. Whereas traditional optimization procedures result in an unique solution point, MO methods provide a set of optimal solutions (Pareto set). The general formulation of a MO problem is expressed by

where represents a decision vector, is the th objective function, is the domain of solutions, and are the equality end inequality constraints, respectively. In order to solve an MO optimization problem, three basic steps have to be taken: dene the objective functions, nd the noninferior solutions and nally choose a solution from this set. The decision maker has to consider the relative importance of the conicting objectives and the nal global noninferior solution depends on his point of view. Noninferiority means that improvements in one objective (e.g., the cost of power losses) is attained only at the cost of some sacrice in the other objective functions (e.g., investment cost or reliability). There are various techniques for generating noninferior solutions. In the paper the -constrained method has been applied [24], [25]. In the -constrained technique, a particular objective function is selected as master objective function (e.g., the th function). The other objective functions, the slave ones, are regarded as new constraints to be complied with. The resulting problem can be formulated as in (2), where represents the upper limit of the th objective

(2) and can be obtained by adding to the initial noninferior value of the -th objective, , the trade-off preference assigned by the . The trade-off preferences can be seen as the complanner, promised value between the conicting objectives. The decision maker on the basis of his experience and/or system operating policies can establish them. IV. OBJECTIVE FUNCTIONS FOR THE MULTIOBJECTIVE OPTIMAL DG SITING AND SIZING The main goal of the proposed planning algorithm is to determine good locations for new generators and their optimal size by minimizing different functions related to the cost of energy losses, the cost of service interruptions, the cost of network upgrading, and the cost of energy purchased. Such objectives should be met subject to the network power ow equations as well as to the limits on the bus voltages, steady state current and short circuit currents (see Appendix A for details on the probabilistic load implemented and the technical constraints). A mathematical expression of the problem is (3) is a power ow solution calculated as function where of the vector , which stores data about the position and the size of the generators. The optimization problem is constrained, nonlinear, with mixed integer variables (due to the discrete size of DG units), and it may be solved with a GA [10], [12]. In the following, each cost function is described in detail under the hypothesis of a linear growth rate of the power demand during the whole study period.

(1)

752

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 2, MAY 2005

A. Cost of Network Upgrading This cost takes into account investments that can be necessary toupgradeadistribution networktofacewith thenaturalgrowthof energy demand or with the appearance of new loads or generators connected to the grid. The objective function to be minimized is of the generic network, with thus represented by the total cost present value taken at the beginning of the whole planning period years. The general expression is the following: of (4) is the number of network nodes, is the number where is the number of branches in the netof substations, work, is the present cost of the th branch, and is the cost of Automatic Sectionalizing Switching Devices (ASSDs) in the network. The cost of every branch is the sum of construction, residual, and management costs transferred to the cash value at the beginning of the planning period by using economical expressions based on the ination rate, the interest rate, and the load growth rate (all of them constant). B. Cost of Purchased Energy In order to allow assessing the more convenient penetration level of DG in a given distribution network, the cost of purand from chasing energy from the transmission grid has been considered. DG In a competitive electricity market, different retail sale rates of the energy produced by a DG unit, that depends on the technology adopted (mini gas turbine, CHP, wind turbine, etc.), and of the energy fed by transmission system have to be considered. By assuming a constant power demand growth rate and by calculating the amount of energy generated per year by each generator on the basis of its power production probability density function, it is possible to assess the energy that a DISCO may buy from both the transmission system andthe DG installed in its network duringthe study period. By resorting to an average value of the energy rate in and the planning period, it is easy to calculate the terms ,opportunelytransferredto thecash valueatthebeginning of the planning period, so that they can be comparable with the other costs of the objective function. The authors are aware that calculating the cost of energy from transmission and DG is not an easy task and that many costs terms cannot be exactly known. Anyway, it should be noticed that more complex cost models can be easily introduced without modifying the general approach proposed by the authors. Equation (5) allows assessing the impact of DG on MV distribution networks and the more convenient penetration level. Indeed, this limit will be reached when the benets, achievable by adding new DG, no longer compensate their relative costs (5) C. Cost of Energy Losses This objective attempts to minimize the total cost of the energy losses arising from line branches. By dividing the whole years into subperiods, the Net Present planning period of

Value (NPV) of the power losses cost for the th branch in the th subperiod can be calculated as follows:

(6) where is the annual power losses cost, and are the branch current at the beginning and at the end of the th subperiod, respectively, represents the duration of the th is the beginning year of the th subperiod and subperiod, is the actualization rate. The total energy losses cost is for each branch and each then obtained as the sum of the subperiod [see Appendix B for more details on (6)]. D. Cost of Energy Not Supplied In order to calculate the cost of energy not supplied the duration of a branch fault is usually divided into two phases: fault location and fault repair [20]. Automatic sectionalizers and reclosers can restrict the area of inuence of a fault, reducing the number of customers affected by long-term interruptions during the fault location phase. In this stage, intentional islanding may be used to supply unfaulted portions of the network automatically separated from the faulted section. The repair stage consists of the time required to isolate the faulted branch, connect any emergency ties and repair the fault. DG, enabling power to be restored to the nodes downstream the sectionalized branch, can lead to signicant reliability improvements [21]. Load ow studies should be performed to check that voltages and currents are within their operative ranges and that DG units have a sufcient probability to pick up the loads in the islanded network. Equation (7) gives the contribution of the th network branch in the th subperiod to the annual cost of energy not supplied for a generic combination of ASSD

(7) where is the branch fault rate (number of faults per year is the branch length (km), is the and km of feeder), and are the cost of the energy not supplied ($/kWh), number of nodes isolated during the fault location and repair is the node power (kW) at the beginstages, respectively, and are the durations of ning of the th subperiod, and dethe fault location and repair stages (h), respectively. pends on the presence of emergency ties, whereas and decreases with the number of ASSD, because the fault location becomes easier and faster. The optimal number and position of ASSD is determined with the algorithm in [20] and [21]. The NPV of the cost of energy not supplied due to a fault in the th branch during the th subperiod is calculated with the following expression:

(8)

CELLI et al.: A MULTIOBJECTIVE EVOLUTIONARY ALGORITHM

753

where is the load growth rate assumed equal for each node. is then obtained as the The cost energy not supplied for each branch and each subperiod [see sum of the Appendix C for more details on (8)].

V. GA IMPLEMENTATION In this paper, a GA optimization technique developed by the authors in a previous work [12] has been used for nding the noninferior solutions of the MO optimization algorithm. The rst important aspect of a correct GA implementation is the coding of the potential solution. If the network structure is xed, all the branches between nodes are known, and the evaluation of the objective functions described above depend only on the size and location of DG units. For this reason each solution can be coded by using a vector, whose size is equal to the number of MV/LV nodes, in which each element contains the information on the presence or not of a DG unit. A binary coding would be sufcient to solve the DG siting problem (1 for presence and 0 for no presence of DG unit), but not the DG sizing one. Therefore, a prexed number (NDG) of generator sizes have been assumed, and each element of the vector solution is represented by means of the following alphabet: no DG located in the node size index of the DG installed in the node

Fig. 1.

Block diagram of the optimization algorithm.

VI. MO ALGORITHM IMPLEMENTATION The block diagram of the described MO optimization algorithm is depicted in Fig. 1. The starting conguration is achieved by minimizing the weighted sum of all the objective functions. Afterwards, the iterative application of the -constrained technique allows to generate the Pareto set, which satises the decision makers requirements [24]. At the beginning of the iterative MO procedure, the planner denes the master objective , whereas the other slave functions are regarded function as constraints. Furthermore, the decision maker has also to for each single choose acceptable trade-off variation levels objective function . Once the setting of these parameters is completed, the procedure can automatically generate a nondominated solution. The master function is optimized with the same GA used for nding the starting solution. Of course, the GA objective function must be modied according to the planners choice: in this phase, the GA looks for those solutions which improve the master function, and simultaneously complies with constraints remaining within the allowed variation margin of deterioration in one or more objective functions. By so doing, a set of potential nondominated solutions can be achieved with a new GA run. The optimization process can be stopped when the solutions are acceptable or it can be restarted with until a satisfactory result is obtained. In different values of particular, the planner can decide that a network characterized by a lower building and maintenance cost to the detriment of energy losses or not supplied energy costs is more suited to his needs (provided that such increase remains within the prexed threshold). Obviously, the nal decision, about which solution from the Pareto set can be advantageously adopted, relies on the decision maker. This feature of the procedure can be very useful in the present scenario, which requires the planner to consider several alternatives for different uncertain futures. The presented optimization procedure permits the denition of the optimal size and allocation of DG in a given network as a compromise among different noninferior solutions produced by the application of the MO methodology. This way of proceeding may be very useful because it allows the planner to be aided by a software tool, able to take into account quickly and precisely all possible combinations in real size cases, but it does leave to

NDG

The implemented GA starts by randomly generating an initial population of possible solutions. For each solution a value of DG penetration is chosen between 0 and a maximum limit, xed by the planner on the ground of economical and technical justications; than, a number of DG units of different sizes is randomly chosen until the total amount of power installed reaches the DG penetration level assigned. At this point, the DG units are randomly located among the network nodes and the objective function is evaluated verifying all the technical constraints; if one of them is violated, the solution is penalized. Regarding the population size, the best results have been found assuming it equal to the number of network nodes. Once the initial population is formed, the genetic operators are repeatedly applied in order to produce the new solutions. In particular, a classical remainder stochastic sampling without replacement scheme has been adopted for the selection operator [26], and a uniform crossover has been chosen by which each vectors element is swapped with probability 0.5. For the mutation operator, all the vector elements are mutated, with a small mutation probability, choosing a different value in the dened alphabet. If one technical constraint is violated or the total amount of DG exceeds the maximum level of DG penetration, the new solution is penalized. Finally, according to the GA steady-state typology, the new population is formed comparing old and new solutions and choosing the best among them. The algorithm stops when the maximum number of generations is reached or when the difference between the objective function value of the best and the worst individuals becomes smaller than a specied value.

754

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 2, MAY 2005

TABLE I COSTS PROGRESSION IN MO ITERATIVE PROCEDURE (CASE STUDY 1)

TABLE II COSTS PROGRESSION IN MO ITERATIVE PROCEDURE (CASE STUDY 2)

the decision maker the control on the process to make the nal decision. VII. RESULTS AND DISCUSSION In order to show the capability of the proposed methodology to solve the problem of the optimal DG allocation, a small portion of a distribution network constituted by 142-MV/LV nodes and two primary substations has been considered. The period taken into consideration for the planning study is 20 years long, with all nodes existing at the beginning of the period. The annual medium active power delivered to the MV nodes is, at the beginning of the this period, about 5MW. For each MV/LV node a constant power demand growth rate of 3% per year has been assumed (this assumption has been made for the sake of clarity but there are no restrictions to dene a power demand growth rate differentiated for each node and for each subperiod). The majority of the branches are of the overhead type, but some buried cables exist. The optimization algorithm can choose different sizes of DG generators within a discrete number of prexed sizes. In the proposed application, generator sizes of 200400600 kW have been adopted. The price of the energy purchased from the wholesale electricity market has been assumed equal to 4.0 /kWh, whereas the price of the energy supplied by DG has been considered equal to 4.5 /kWh. This low price may be considered acceptable only for the current cheapest energy sources (e.g., large scale wind farms) and has been adopted in order to stress the methodology on a small example. It should be highlighted that, in presence of a liberalized electricity market, different retail sales rate of the energy produced by a DG unit should be considered. These retail sales depend on the technology adopted (mini gas turbine, CHP, wind turbine, etc.), the regulatory actions and the willingness to harness renewables. The application of the GA to minimize the generalized cost of the network (see Section IV) has allowed nding the starting conguration. The global cost of the network during the assigned study period is equal to k$27 008.1 (see Tables I and II) without any operating DG. Such a high generalized network cost is due to the signicant growth rate of the demand, which requires the enforcement of a large number of branches. The attempt to minimize the global cost has led to a solution with many lines close to their maximum capacity and for this reason the cost of the energy losses counts for a signicant percentage of the generalized cost of the network. The use of DG as an electric supply option can reduce both costs.

Very often the planner needs more alternatives to evaluate and sometimes he can prefer to reduce the cost of losses instead of improving service quality, depending on strategic decisions, regulatory directives regarding the electric service, and budget restrictions. As showed in the following examples, the proposed MO optimization process permits of nding out alternative congurations, characterized by different costs for each single function constituent of the global cost. In each optimization stage the MO algorithm looks for alternative solutions, which improve the master function to the detriment of the slave functions. Two different cases have been investigated. In the rst case has been regarded as the study, the power losses cost master objective function. The value assumed by the cost of in the initial network conguration is the power losses equal to k$1 076.5. Two consecutive steps of iteration have been run. In the rst iteration, a deterioration margin of the slave functions, necessary to allow the GA a more exhaustive exploration in the space of solution, has been admitted. By so doing, the new optimized conguration permits reducing the in percentage of the 23% against an power losses cost increase in the global cost. This losses reduction is obtained resorting to a DG penetration level of 13.64% (DG% is the ratio between the DG capacity and the power of load). In ,a the second iteration, by admitting a larger variation of new DG arrangement has been found by the GA. The optimal network conguration, corresponding to the last iteration, is shown in Fig. 2. In the new optimal solution the cost of decreases from the value of k$829.0 to the value of k$656.5. The penetration level of DG increases from the value of the 13.64% to the value of 44.34% and consequently the cost of the purchased energy is higher. This kind of optimization can be very useful if the decision maker is committed to reduce losses and to improve network performance maintaining investments to a reasonable low level. This feature may be very useful in case the revenue recovered by DISCO is not based on the asset value but also on network performance (e.g., on the level of losses or reliability). In this scenario, the objective of the distribution planner is not the reduction of global costs but the individuation of a trade-off solution that allows both cost reduction and performance maximization. In the second case study, the cost of service interruptions has been regarded as the master objective function. In this case the planner aims at reducing the number and the duration

CELLI et al.: A MULTIOBJECTIVE EVOLUTIONARY ALGORITHM

755

Fig. 2.

DG units as positioned by the GA (Case Study 1, MO iteration #2).

TABLE III DATA USED FOR RELIABILITY CALCULATION

Fig. 3. DG units as positioned by the GA (Case Study 2, MO iteration #2). Light and heavy gray areas identify the intentional islands formed during the fault location and repair stages, respectively, for the fault indicated with the black arrow. Each area is accepted if the probability that generation exceeds load demand is greater than 30%. TABLE IV RELIABILITY INDEXES IN THE NETWORK DEPICTED IN FIG. 3

of service interruptions by positioning DG in suited locations (Table III reports data used for reliability calculations). Even though many standards and almost all the distributors do not generally allow resorting to intentional islanding operation, in order to emphasize the effect of DG it has been hypothesized that this practice can take place [20], [21]. Indeed, among the many new features of new active networks, the possibility of using DG to supply loads in self-sustaining islands seems to have a real interest for the local distribution companies. In [21] Celli et al. have pointed out that the practice of intentional islanding can be really useful for those nodes that suffer for low service continuity (e.g., nodes that cannot use alternative energy routes during faults in the main feeder), provided that DG could feed the island and network automation is optimally located. In the proposed example, the capability of optimizing the location of DG has been advantageously used to nd a network arrangement able to give the customers a much more reliable service avoiding the construction of new emergency ties. The starting network conguration is equal to the previous cost has the value of k$1 226.6. case study, where the With the rst optimization step this value is reduced to k$893.5 (see Table II). A further optimization permits reducing up to k$801.3 thanks to a new allocation of DG (see Fig. 3) and the purchase of more energy from DG. It is worth noticing that in this case generators are located at the end of long and heavy loaded lateral edges to serve as back up energy sources

during upstream faults. Global benets on ENS and cost of not supplied energy are clearly recognizable, but benets are much more signicant for those customers that suffer for poor quality due to their position in the network. For example, nodes 1 and 2 in Fig. 3, due to the coordination between DG and network automation, drastically reduce their number and duration of long interruptions (see Table IV). It can be easily noticed that system indexes cannot capture the importance of intentional islanding for customers seeking for premium power contracts. Finally, it is prominent to observe that this amelioration can only be achieved by accepting a major generalized cost of the network and this fact justies the adoption of MO approaches, that allows the planner to stress some terms of the objective function in order to get some specic results (e.g., reliability improvements under regulatory actions or market pressure). VIII. CONCLUSION Nowadays, a number of impediments and barriers still inhibit an increased market penetration of DG. The high costs and uncertain performances of many DG technologies, the lack of uniform standards and communication protocols, the architecture of the distribution system are examples of the most common barriers to the DG development. Anyway, DG is destined to play

756

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 2, MAY 2005

an important role in the distribution system of the future. Distribution engineers need new planning tools to maximize benets in the new uncertain scenario. According to these assumptions, an algorithm for the optimal allocation of DG in a given network is proposed. The procedure, based on the application of GA and MO, allows the planner to drive the solution toward his particular requirements. In the context of a liberalized energy market, it can be used to nd the most valuable sites to exploit and evaluate any additional credits DISCO might offer if the DG is placed in the appropriate location to have real benets for the network [7], [8]. Further studies will improve the proposed MO methodology taking into account the liberalized energy market, the interaction with the transmission grid, the market of ancillary services, DSM, etc. In addition, the consideration of the power quality aspects in the planning of DG siting and sizing is compulsory. APPENDIX A. Load Flow Calculation and Technical Constraints In order to consider the random behavior of DG and loads, a simplied probabilistic load ow (PLF) has been developed and implemented. This procedure takes into account the pdf of both loads and generators. Correlation among DG units, between generators and loads and also among loads have been considered with the assumption that such correlation is linear (no dispatch action at the distribution level) [27], [28]. Once the currents owing in each branch and the voltage of each node are calculated as probabilistic quantities, they can be used to design the network so that all the technical constraints are complied with [29]. In particular, only those congurations having a reasonable probability to verify the constraints on the maximum and minimum allowable node voltage can be accepted. Known the pdf of the voltage module in each node, the probability to have a voltage drop or an overvoltage greater than the maximum allowable limit is evaluated. Then, this value is compared with an acceptable value of probability dened by the planner and, if necessary, actions are taken to bring back the voltage within its limits (e.g., network enforcement). A similar approach has been followed for network sizing. Firstly, the planner denes an acceptable probability of overload occur. Then, from the pdf of the branch current the rence current value that has a probability of occurrence greater than is derived. By comparing, for each network branch, the current with the corresponding branch capacity, it is straightforward that those conductors whose capacity is smaller than should be reinforced [29]. B. Actualization Factor for Power Losses The cost of power losses for the th branch in one year of the th subperiod is evaluated assuming a constant branch current during the year, equal to the value of the current at the end of the year. Then, this cost has to be actualized in order to obtain its NPV. By summing this costs for each year of the subperiod, the whole NPV of power losses cost for the th branch in the th subperiod is assessed.

Supposing a linear growth of the branch current in the th can be subperiod, the current value at the end of each year easily evaluated by (9) is the duration where is the th year of the th subperiod, in years of the th subperiod, and are the current of the th branch evaluated, respectively, at the beginning and at the end of the th subperiod, by means of adequate load ows. Equation (9) can be rewritten as follows: (10) By applying the actualization rate , function of the ination rate and the interest rate, the NPV of the power losses cost for the th branch in the th subperiod can be written as

(11) where is a factor that depends on the unitary cost of the power losses and the resistance of the th branch, and is the beginning year of the th subperiod. C. Actualization Factor for Energy Not Supplied The cost of energy not supplied due to a fault on the th branch in the th subperiod is evaluated referring to the power of the isolated nodes. Thus, it is necessary to calculate the total power absorbed by these nodes in the whole subperiod, and to actualize the related disruption cost in order to assess the NPV of the cost of energy not supplied. Supposing a linear growth of the power absorbed by the th can be treated as the node, the power demand growth rate angular coefcient of the power growth line, expressed in p.u. . In of the power node at the beginning of the th subperiod this case the power at the end of the th year of the subperiod can be expressed by (12) The NPV of the cost of the energy not supplied to the th node in the th subperiod due to a fault in the th branch can be evaluated as follows:

(13)

CELLI et al.: A MULTIOBJECTIVE EVOLUTIONARY ALGORITHM

757

where is a factor that depends on the unitary cost of the energy not supplied, the fault rate and the length of the th branch. Equation (8) is nally obtained by summing this cost for all the nodes that remain isolated due to a fault on the th branch, and assuming a power demand growth rate equal for each node. ACKNOWLEDGMENT This paper is based on a contribution presented by the authors at the IEEE PowerTech Conference, Bologna, Italy, June 2326, 2003. REFERENCES
[1] Impact of Increasing Contribution of Dispersed Generation on the Power SystemFinal Report,, CIGRE WG 37-23, Sep. 1998. [2] Dispersed generationPreliminary report, presented at the CIRED Conf., Nice, France, Jun. 25, 1999. CIRED WG04. [3] H. L. Willis and W. G. Scott, Distributed Power Generation. New York: Marcel Dekker, 2000. [4] P. P. Barker and R. W. de Mello, Determining the impact of distributed generation on power systems: Part 1Radial distribution systems, in Proc. IEEE Power Eng. Soc. Summer Meeting, vol. 3, Seattle, WA, Jul. 1620, 2000, pp. 16451656. [5] N. Jenkins, R. Allan, P. Crossley, D. Kirschen, and G. Strbac, Embedded Generation. London, U.K.: IEE, 2000. [6] T. E. McDermott and R. C. Dugan, PQ, reliability and DG, IEEE Ind. Appl. Mag., vol. 9, no. 5, pp. 1723, Sep.Oct. 2003. [7] R. C. Dugan and S. K. Price, Issues for distributed generation in the US, in Proc. IEEE Power Eng. Soc. Winter Meeting, vol. 1, New York, Jan. 2002, pp. 121126. [8] R. C. Dugan, T. E. McDermott, and G. J. Ball, Planning for distributed generation, IEEE Ind. Appl. Mag., vol. 7, no. 2, pp. 8088, Apr. 2001. [9] G. W. Ault and J. R. McDonald, Planning for distributed generation within distribution networks in restructured electricity markets, IEEE Power Eng. Rev., vol. 20, no. 2, pp. 5254, Feb. 2000. [10] G. Carpinelli, G. Celli, F. Pilo, and A. Russo, Distributed generation siting and sizing under uncertainty, in Proc. IEEE Powertech Conf., vol. 4, Porto, Portugal, Sep. 1013, 2001, pp. 376401. , Distributed generation planning under uncertainty including [11] power quality issues, ETEP, vol. 13, no. 6, pp. 381389, Nov./Dec. 2003. [12] G. Celli and F. Pilo, Optimal distributed generation allocation in MV distribution networks, in Proc. IEEE PICA Conf., Sydney, NSW, Australia, May 2024, 2001, pp. 8186. [13] I. Bouwmans. Distributed NetworksThe Grid and the Net Compared. [Online]. Available: www.inem.tbm.tudelft.nl/ivo [14] V. Roberts, A. Beddoes, A. Colinson, and F. van Overbeeke, Active networks for the accommodation of dispersed generation, in CIRED Conf., Barcelona, Spain, May 1215, 2003, pp. 16. Session 4, Paper 51. [15] Review of Technical Options and Constraints for Integration of Distributed Generation in Electricity Networks [Online]. Available: http://www.sustelnet.net [16] Grid of the Future White Paper on Interconnection and Controls for Reliable, Large Scale Integration of Distributed Energy Resources [Online]. Available: http://certs.lbl.gov [17] T. Bopp et al., Commercial and technical integration of distributed generation into distribution network, presented at the CIRED Conf., Barcelona, Spain, May 1215, 2003, pp. 14. Session 4, Paper 45. [18] I. Chilvers, N. Jenkins, and P. Crossley, Development of distribution network protection schemes to maximize the connection of distributed generation, presented at the CIRED Conf., Barcelona, Spain, May 1215, 2003, pp. 14. Session 4, Paper 46. [19] C. A. Coello Coello, An updated survey of evolutionary multiobjective optimization techniques: State of the art and future trends, in Proc. IEEE Congr. Evol. Comput., 1999, pp. 313. [20] G. Celli and F. Pilo, Optimal sectionalizing switches allocation in distribution networks, IEEE Trans. Power Del., vol. 14, no. 3, pp. 11671172, Jul. 1999. [21] F. Pilo, G. Celli, and S. Mocci, Improvement of reliability in active networks with intentional islanding, in DRPT Conf., vol. 2, Honk Kong, Apr. 58, 2004, pp. 474479. Paper 117. [22] Directive 2003/54/EU (2003, Jun. 26). [Online]. Available: http://europa.eu.int/eur-lex

[23] G. Strbac and N. Jenkins, Calculation of cost and benets to the distribution network of embedded generation, presented at the Inst. Elect. Eng. Colloq. Econ. Embedded Gen., Oct. 29, 1998, pp. 6/16/13. Ref. no. 1998/512. [24] Y. T. Hsiao and C. Y. Chien, Optimization of capacitor allocation using an interactive trade off method, Proc. Inst. Elect. Eng., Gen., Transm., Distrib., vol. 148, no. 4, pp. 371374, Jul. 2001. [25] A. Berizzi, C. Bovo, and P. Marannino, The surrogate worth trade-off analysis for power system operation in electricity markets, in Proc. IEEE Power Eng. Soc. Summer Meeting, vol. 2, Vancouver, BC, Canada, Jul. 1519, 2001, pp. 10341039. [26] D. E. Goldberg, Genetic Algorithms in Search, Optimization & Machine Learning. Reading, MA: Addison Wesley, 1989. [27] A. Dimitrovski and R. Ackovski, Probabilistic load ow in radial distribution networks, in Proc. T&D Conf., 1996, pp. 102107. [28] R. N. Allan and M. R. G. Al-Shakarchi, Linear dependence between nodal powers in probabilistic a.c. load ow, Proc. IEE, vol. 124, no. 6, pp. 529534, Jun. 1977. [29] G. Celli, R. Cicoria, S. Mocci, and F. Pilo, Probabilistic optimization of MV distribution network in presence of distributed generation, in Proc. PSCC Conf., Sevilla, Spain, Jun. 2428, 2002, pp. 17. paper S11-1.

Gianni Celli (M98) was born in Cagliari, Italy, in 1969. He received the electrical engineering degree from the University of Cagliari in 1994. He has been an Assistant Professor of Power System with the University of Cagliari since 1997. Current research interests are in the eld of MV distribution network planning optimization, power quality, and neural networks. He is the author of several papers published on international journals or presented in various international conferences. He is an AEI member.

Emilio Ghiani was born in Oristano, Italy, in 1973. He received the electrical engineering degree from the University of Cagliari, Cagliari, Italy, in 1999. He has been an Assistant Professor of Power System with the University of Cagliari since 2003. Current research interests are in the eld of MV distribution network planning optimization and power quality. He is an AEI member.

Susanna Mocci (M04) was born in Cagliari, Italy, in 1973. She received the electrical engineering degree from the University of Cagliari in 2001, where she currently is working toward the Ph.D. degree. Her research activity is focused on DG and distribution system. She is an AEI member.

Fabrizio Pilo (M97) was born in Sassari, Italy, in 1966. He received the electrical engineering degree from the University of Cagliari, Cagliari, Italy, in 1992 and received the Ph.D. degree from the University of Pisa, Pisa, Italy, in 1998. Since 2001, he has been an Associate Professor of Electrical Power Systems at the University of Cagliari. His research activity is focused on electrical power systems, network planning and optimization, and distributed generation. He is author of several papers published on international journals or presented in various international conferences. He is an AEI member. He is Special Reporter of the CIRED S5 and invited expert at the CIGRE SC6.

Vous aimerez peut-être aussi