1 Implementation of Center d Voltage and Var Optimization in Distribution Management System Xiaoming Feng, William Peterson, Fang Yang, Gamini M. Wickramasekara, John Finney Abstract--This paper presents the implementation of an online voltage and var optimization (VVO) application that runs at the control center for distribution systems. The innovative VVO application combines state of the art optimization technology with detailed and accurate modeling of distribution systems. The multi-phase, unbalanced system model can have radial or meshed circuits, single or multiple sources, wye and/or delta connected transformers, grounded or ungrounded, voltage dependent load models, ganged or unganged controls of capacitor banks and voltage regulators. This application is capable of optimizing large and complex networks with online application speed. It is designed for smart grid application to optimize distribution energy delivery efficiency by minimizing the energy loss and manage system demand. Index Terms Distribution management system, energy efficiency, multi-phase, multi-source, unbalanced, meshed, reactive compensation devices, utility control center, var compensation, voltage and var optimization (VVO), voltage regulating devices I I. INTRODUCTION n a world strapped for energy and in need for significant reduction in carbon emission, energy efficiency and demand management have become key strategies for electric power grid operators all over the world to make the best use of the energy, optimize utilization of transmission and distribution assets, and meet growing demand. Distribution systems, which historically accounted for about 4~5% of the total energy losses from power plants to end users, is the last frontier for energy efficiency and demand management technology. Most of the energy loss occurring on the distribution system is the ohmic loss resulting from the electric current flowing through conductions made from aluminum or copper. The amount of loss is proportional to the product of the conductor resistance and the square of the current magnitude. Energy loss can be reduced, therefore, either by reducing resistance or the current magnitude or both. The resistance of a conductor is determined by the resistivity of the conductor material, by its cross sectional area, and by its length, none of which can All authors are affiliated with ABB Inc., Raleigh, North Carolina, 27606. be changed easily in existing distribution networks. However, the current magnitude can be reduced by eliminating unnecessary current flows in the distribution network. For any conductor in a distribution network, the current flowing through it can be decomposed into two components: active and reactive. The active component is in phase with the terminal voltage phasor angle and delivers the active energy that does the real work for a utility. The reactive component is orthogonal in phase to the terminal voltage phasor angle and delivers the reactive power that does not do any real work, but uses nevertheless the energy delivery capacity of the conductors. Reactive power (var) compensation devices are designed to reduce or eliminate the unproductive component of the current and thus energy loss. Reactive compensation devices, such as capacitor banks, are used to reduce the reactive power flows through out the distribution network. The capacitor banks may be located in the substation or on the feeders. The voltage profile on the feeders can also affect the current distribution, although indirectly and to a smaller extent, affect energy loss. On the other hand, the voltage profile could have significant influence on the system demand for the system that has considerable amount of voltage dependent loads, such as constant impedance loads. Voltage regulating devices include substation transformers that have on load tap changers and special transformers with tap changers called voltage regulators at various locations on the feeders. By adjusting the tap settings of these devices, the voltage profiles on the feeders can be controlled to varying degrees. Historically, the voltage and var control devices are regulated in accordance with locally available measurements of, for example, voltage or current. On a feeder with multiple voltage regulation and var compensation devices, each device is controlled independently, without regard for the resulting consequences of actions taken by other control devices. This practice yields inconsistent control actions and sub-optimal results. To achieve optimal result, a coordinated approach is preferred. The coordination could be done centrally using a substation automation system or a distribution management system. Two main obstacles prevented the adoption of coordinated control: 1) affordable two-way communication infrastructure, and 2) robust voltage and var optimization algorithm to handle large and complex real distribution systems. 978-1-4244-6547-7/10/$26.00 2010 IEEE
2 The accelerated adoption of substation automation (SA), feeder automation (FA) technology, and the wide-spread deployment of advanced metering infrastructure (AMI) have over the last few years laid the foundations for a centralized control approach, by providing the necessary sensor, actuator, and reliable two-way communications between the field and the distribution system control center. Until recently, however, VVO technologies have not been mature enough to optimize large and complex distribution systems with satisfactory performance in solution quality, robustness, and speed. The search for VVO methodologies has been going on for several decades. References [1]-[7] are representatives of using analytical techniques to solve the problem, in which the distribution system is generally modeled as the radial system, the impact of the detailed and accurate system component models on the solution is not taken into account, and the optimization problem is simplified to make it easy to solve. In the last decade, some other techniques have been attempted to solve the problem, including the rule based techniques [8]- [14] and Meta-heuristic techniques [15]-[18] (e.g., genetic algorithms, simulated annealing, particle swarming, etc). These approaches can avoid the modeling complexity. However, they are limited in solving small scale problems and in off-line applications where online performance is not required. This paper begins with a review of the technical challenges for voltage and var optimization for the distribution systems. It describes the special requirements for modeling distribution systems with unbalanced construction and loading conditions, especially for meshed network with multiple sources. It then presents an implementation of a centralized voltage and var optimization application. A general formulation of the centralized VVO is provided, where the problem is cast as mixed integer quadratic programming for a var optimization sub-problem and iterative mixed integer quadratic programming for a voltage regulation sub-problem. Representative results of the centralized VVO using real utility circuits with thousand nodes are provided to illustrate the effectiveness of this implementation. II. TECHNICAL CHALLENGES The objective of VVO is to minimize the energy loss on a distribution circuit or the total MW demand supplied from a substation. The controls are the switching status of the capacitor banks and the tap settings of voltage regulation devices. Two main obstacles lie ahead of the coordinated voltage and var optimization. 1) Accurate modeling of the distribution system s behavior under any control setting 2) Efficient and robust search algorithm for optimization to provide discrete solution for capacitor bank status and taps of voltage regulating devices. A good VVO solution approach should be able to handle large, real distribution systems that have any combinations of the following characteristics: Multi-phase unbalanced construction and loading Radial or meshed systems Single or multiple sources Various transformer connections (Y/Y, Y/, /Y, /, grounded or ungrounded) Various load connections (Y or ) Various combination of voltage dependent load models Ganged or un-ganged control for capacitor banks and voltage regulating devices. VVO in essence is a non-linear combinatorial optimization problem with the following characteristics: o Integer decision variables both the switching status of capacitor banks and the tap position of regulation transformers are integer variables o Nonlinear objective being an implicit function of decision variables energy loss or system demand are implicit functions of the controls o High dimension nonlinear constraints power flow equations numbering in the thousands in the multiphase system model o o Non-convex objective function and solution set High dimension search space with un-ganged control, the number of control variables could double or triple. For a distribution circuits with 5 capacitor banks, each of which has two status (on and off), and 5 voltage regulating devices, each of which has 32 tap settings, the total number of combinations of controls for ganged operation is 2 5 32 5, or for un-ganged operation, 2 15 32 15. It is well known among operation research theoretician and practitioners that for such mixed integer nonlinear, non convex (MINLP-NC) problems, it is very difficult to develop efficient search algorithms. A good algorithm is the one that delivers optimal or very near optimal solution efficiently, which is a very essential attribute for online applications. Since a certain amount of computational resource (CPU time) is needed to evaluate the loss or demand for a single specific control solution (a single functional evaluation), an algorithm that requires fewer number of functional evaluation in order to find the optimal solution is generally regarded as more efficient than one that requires more functional evaluations to achieve the same objective. In the case of VVO, a single function evaluation involves solving a set of non-linear equations, the unbalanced load flow, with several thousand state variables. III. CONTROL CENTER BASED VVO This paper presents a control center based VVO that has been recently developed and implemented by authors. This new generation of VVO is capable of optimizing very large and complex unbalanced meshed distribution networks with online application speed. It combines accurate distribution system modeling (a general purpose unbalanced load flow model) and state of the art optimization techniques built on mixed integer quadratic programming to overcome the technical challenges for coordinated VVO. A. General Problem Definition for VVO The objective of VVO is to minimize the weighted sum of energy loss MW load Voltage violation Current
3 violation, subject to a variety of engineering constraints, such as: o Power flow equations (multi-phase, multi-source, unbalanced, meshed system) o Voltage constraints (phase to neutral or phase to phase) o Current constraints (cables, overhead lines, transformers, neutral, grounding resistance) o Tap change constraints (operation ranges) o Shunt capacitor change constraints (operation ranges) The control variables for the optimization include: o Switchable shunts capacitor banks (ganged or unganged) o lable taps of transformer/voltage regulators (ganged or un-ganged) B. Solution Approach The VVO problem is formulated as a sequence of capacitor (var) only optimization and voltage regulation (voltage) only optimization tasks. The var optimization task is formulated as a mixed integer quadratic program. The voltage optimization problem is formulated as a sequence of LP problems [19]. C. Distribution Network Model Detailed and realistic network modeling is essential to the proper functioning of VVO. Real distribution networks do not change from unbalanced operation to balanced operation, or from a meshed network to radial network so that a simplified network model can be used. The model used for VVO must fit the operation reality, not the other way around. For VVO, phase based models are used to represent every network component. Loads/capacitor banks can be delta or wye connected. Transformers can be connected in various delta/wye and various secondary leading/lagging configurations with/without ground resistance, with primary or secondary regulation capability. Figure 1- Figure 4 provide representative model examples of various distribution network components including distribution feeders, loads, capacitor banks and transformers. I ~ La I ~ Lb I ~ Lc ab bc ca Figure 2 : -connected load I ~ Ca I ~ Cb I ~ Cc bn jb a an jbb jbc - cn Figure 3 : Y-connected capacitor bank I ~ A A B C I ~ B I ~ C ~ Y ~ Y ~ Y Figure 4 : Y/Y connected transformer (one of many possible connections VVO can model) D. Implementation Environment The mixed integer quadratic programming based VVO was implemented in a distribution management system (DMS). Shown in Figure 5 is the operator GUI of the DMS, from which an operator can initiate a variety of integrated applications, including load allocation, balanced power flow, unbalanced load flow, outage management, voltage and var optimization, and others. t:1 I ~ a a I ~ - b b I ~ c c k Node ak I ~ ak bk I ~ bk ck I ~ ck I ~ a I ~ b I ~ c z ~ aa z ~ bb z ~ cc } z ~ ab z ~ ac } z ~ bc Node l al I ~ al bl I ~ bl cl I ~ cl 1 1 I ~ abcsk [ Y abcs ] [ Y abcs ] I ~ abcsl 2 2 -- --------------------------- - - - -- - - --- - ----- Figure 1 : Three-wire model of three-phase four-wire construction Figure 5 : DMS GUI
4 E. Defining features In the current VVO implementation, both voltage and var controls can be ganged or un-ganged. Voltage constraints are enforced for each individual phase, using phase to ground or phase to phase voltage, depending on the connection type of the load. Table 1 is a concise summary of the features of the proposed VVO in contrast to prior art. Table 1 : Comparison of VVO capability with prior art Prior Art Single phase equivalent model Balanced load Single source Radial system Ganged control Academic system size Offline performance Heuristic Proposed VVO Capability Multi-phase, unbalanced Model Unbalanced load Multi-source Meshed system Ganged or un-ganged control Real utility system size Online performance MIP, MIQP IV. PRELIMINARY EXPERIENCES The VVO implementation has been tested on various utilities distribution networks using a general purpose office desktop computer. The sizes of the test systems range from 1,600 to 7,800 nodes and 1,600 to 8,100 branches per circuit. Optimization improved the loss from 2.5 percent to 67 percent and demand reduction from 1.4 percent to 5.8 percent. The algorithm and its implementation is very efficient, it finished most the optimization in less than 10 seconds. It is able to converge with 5~8 iterations for voltage regulation optimization with fifteen un-ganged tap controls. The key dimensions of the tested system circuits are provided in Table 2, showing for each circuit the, numbers of circuit total components, feeders, nodes, loads, lines, capacitor banks, regulating transformers. Table 3 shows the approximate number of nodes, loads, and branches of these test circuits. These parameters are indicators of the size of the nonlinear equations that unbalanced load flow needs to solve. Table 2 : Test system dimension Comp. Feeder Node Load Line Cap. Bank Reg. Xfrm 1 1760 4 673 278 699 7 1 2 2455 3 946 431 966 8 1 3 4869 4 1859 865 1916 6 2 4 2327 6 874 308 919 8 1 5 2167 5 820 285 853 6 0 6 1406 4 533 176 560 4 1 7 6987 3 2628 1384 2708 7 2 8 4512 4 1700 858 1759 8 2 Comp. Component, Cap. Capacitor Reg. Xfrm Regulating Transformer Summarized here are representative test results for var only optimization, voltage regulation optimization, and combined voltage and var optimization. For var optimization, Table 4 and Table 5 provide each test case with the information including specific load allocation (LA) level, percentage of constant power load, capacitor control type (ganged or unganged control), the demand and loss before and after optimization, the reduction in demand and loss brought by the optimization. Note that the amount of loss reductions depends on the initial control setting of the controls. Obviously, if the loss optimization starts from an optimal solution, there is little room, if any, to further reduce the loss. Table 3 : Test system load flow dimension Max Nodes Max Loads Max Branches 1 2019 834 2097 2 2838 1293 2898 3 5577 2595 5748 4 2622 924 2757 5 2460 855 2559 6 1599 528 1680 7 7884 4152 8124 8 5100 2574 5277 Table 4 : case results for var only optimization Load Level Constant Power Load Type 1 20% 100% 6605.0 36.5 Ganged 2 20% 100% 8341.1 98.5 Ganged 3 20% 100% 9614.8 91.6 Ganged 4 20% 100% 2941.1 139.5 Ganged 5 20% 100% 2629.6 121.9 Ganged Table 5 : Optimized results for var only optimization Reduction Reduction Reduction 1 6599.1 30.5 5.9 5.9 16.3% 2 8313.4 70.7 27.8 27.8 28.2% 3 9612.3 89.0 2.5 2.5 2.8% 4 2862.3 60.6 78.8 78.9 56.5% 5 2547.2 39.5 82.4 82.4 67.6% Optimized For voltage regulator optimization, Table 6 and Table 7 provide each test case with the information including specific load allocation (LA) level, percentage of constant power load, voltage regulating transformer control type (ganged or unganged), the demand and objective function value before and after optimization, the reduction in demand and objective brought by the optimization. The objective column may have a different value depending if there is any voltage violations, the objective value is the total demand plus and voltage violations weighted by a large penalty factor. After the voltage regulation optimization, both the voltage violations and the total demand have been reduced. It can be seen that when the load is 100% constant power, the effect of voltage regulation optimization is the elimination of the voltage
5 violations and small reduction in the loss. When the percentage of constant impedance load is significant, voltage regulation optimization not only reduces the voltage violations, but also significantly reduces the total demand. Table 6 : case results for voltage regulator only optimization Load Level CPL Objective Type 1 30% 50% 10000.5 10000.4 Unganged 2 30% 50% 12912.8 14748.4 Unganged 3 30% 50% 14618.2 14618.2 Unganged 4 30% 50% 4627.8 24808.1 Unganged 6 45% 100% 3784.1 7152.7 Unganged 8 30% 50% 15567.0 16319.8 Unganged CPL-Constant Power Load Table 7 : Optimized results for voltage regulator only optimization Obj.. Obj. Obj. 1 9548.4 9548.4 452.1 452.0 4.5% 4.5% 2 12254 12254 659.2 2495 5.1% 17% 3 14417 14417 201.0 201.0 1.4% 1.4% 4 4571.6 18072 56.2 6736 1.2% 27% 6 3784.8 3784.8-0.8 3368 0.0% 47% 8 14909 15108 657.8 1212 4.2% 7.4% -Optimized, Obj.- Objective, -Reduction For combined capacitor bank and voltage regulator optimization, Table 8 and Table 9 provide each test case with the information including specific load allocation (LA) level, percentage of constant power load (CPL), capacitor and voltage regulator control type, the demand and loss before and after optimization, the reduction in demand and loss brought by the optimization, as well as the CUP time. The results show that both the total demand and loss are reduced appreciable percentages. Table 8 : case results for voltage and var optimization LA CPL (Kw) Cap. Xfrm. 1 30% 50% 10000.5 74.6 Gang Ungang 2 30% 50% 12912.8 184.4 Gang Ungang 8 30% 50% 15567.0 149.2 Gang Ungang CPL - Constant Power Load, Cap. Capacitor, Xfrm. - Transformer Table 9 : Optimized results for voltage and var optimization. 1 9419.1 69.2 581.3 5.4 5.8% 7.2% 2 12728 162.2 184.4 22.3 1.4% 12% 8 14925 144.1 641.9 5.1 4.1% 3.4% -Optimized, -Reduction V. CONCLUSIONS This paper describes the implementation of a state of the art voltage and var optimization process that can work with multisource, multi-phase, unbalanced, meshed distribution systems. The centralized VVO method has been implemented to work with the comprehensive network modeling capabilities of the unbalanced load flow (UBLF) within a DMS platform. The implementation has been tested with real utility distribution circuits with varying degree of unbalance and model complexity. The implementation demonstrates the feasibility and effectiveness of centralized VVO. VI. REFERENCES [1] J. J. Grainger and S. Civanlar, "Volt/Var on Distribution Systems with Lateral Branches Using Shunt Capacitors and Voltage Regulators-Part 1: The Overall Problem, IEEE Trans. Power Apparatus and Systems, vol. PAS-104, no.11, pp. 3278-3283, Nov. 1985. [2] S. Civanlar and J. J. Grainger, "Volt/Var on Distribution Systems with Lateral Branches Using Shunt Capacitors and Voltage Regulators-Part 2: The Solution Method, " IEEE Trans. Power Apparatus and Systems, vol. PAS-104, no.11, pp.3284-3290, Nov. 1985. [3] R. Baldick and F. F. Wu, "Efficient Integer Optimization Algorithm for Optimal Coordination of Capacitors and Regulators, IEEE Trans. Power Systems, vol.5 no.3, pp.805-812, Aug. 1990. [4] F.C. Lu and Y.Y. Hsu, Reactive Power/Voltage in a Distribution Substation Using Dynamic Programming, IEE Proceedings- Generation, Transmission and Distribution, vol. 142, no. 6, pp. 639-644, November 1995. [5] F.C. Lu and Y.Y. Hsu, Fuzzy Dynamic Programming Approach to Reactive Power/Voltage in a Distribution Substation, IEEE Transactions on Power Systems, vol. 12, no. 2, pp. 681-688, May 1997. [6] I. Roytelman, B. K. Wee, and R. L. Lugtu, and T. M. Kulars, "Volt/Var Algorithm for Modern Distribution Management System," IEEE Trans. Power Systems, vol.10, no.3, pp.1454-1460, Aug. 1995. [7] I. Roytelman, B. K. Wee, and R. L. Lugtu, T. M. Kulars, and T. Brossart, Pilot Project to Estimate the Generalized Volt/Var Effectiveness, IEEE Trans. Power Systems, vol. 13, no.3, pp.864-869, Aug. 1998. [8] S. J. Cheng, O. P. Malik, and G. S. Hope, An Expert System for Voltage and Reactive Power of a Power System, IEEE Trans. Power Systems, vol. 3, no.4, pp.1449-1455, Nov. 1988. [9] M.M.A. Salama and A.Y. Chikhani, An Expert System for Reactive Power of a Distribution System- Part 1: System Configuration, IEEE Transactions on Power Delivery, vol. 7, no. 2, pp 940-945, April 1992. [10] J.R.P-R. Laframbiose, G. Ferland, A.Y. Chikhani and M.M.A. Salama, An Expert System for Reactive Power of a Distribution System- Part 2: System Implementation, IEEE Transactions on Power Systems, vol. 10, no. 3, pp. 1433-1441, August 1995. [11] G. Ramakrishna and N.D. Rao, A Fuzzy Logic Framework for of Switched Capacitors in Distribution Systems, in Proc. Canadian Conference on Electrical & Computer Engineering, Montreal, Canada, September 1995, pp. 676-679. [12] G. Ramakrishna and N.D. Rao, Fuzzy Inference System to Assist the Operator in Reactive Power in Distribution Systems, IEE Proceedings- Generation, Transmission and Distribution, vol. 145, no. 2, pp. 133-138, March 1998. [13] V. Miranda and P. Calisto, A Fuzzy Inference System to Voltage/Var in DMS- Distribution Management System, in Proc. 14 th Power Systems Computations Conference, Sevilla, Spain, June 2002. [14] Y.Y. Hsu and F.C. Lu, A Combined Artificial Neural Network-Fuzzy Dynamic Programming Approach to Reactive Power/Voltage in a Distribution Substation, IEEE Transactions on Power Systems, vol. 13, no. 4, pp. 1265-1271, November 1998. [15] R.H. Liang and Y.S. Wang, Fuzzy-d Reactive Power and Voltage in a Distribution System, IEEE Transactions on Power Delivery, vol. 18, no. 2, pp. 610-618, April 2003. [16] Z. Hu, X. Wang, H. Chen and G.A. Taylor, Volt/Var in Distribution Systems Using a Time-Interval d Approach, IEE
6 Proceedings- Generation, Transmission and Distribution, vol. 150, no. 5, pp. 548-554, September 2003. [17] T. Niknam, A.M. Ranjbar and A.R. Shirani, Impact of Distributed Generation on Volt/Var in Distribution Networks, in Proc. IEEE PowerTech Conference, Bologna, Italy, June 2003. [18] J. Olamaie and T. Niknam, Daily Volt/Var in Distribution Networks with Regard to DGs: A Comparison of Evolutionary Methods, in Proc. IEEE Power India Conference, New Delhi, India, April 2006. [19] X. Feng, W. Peterson, F. Yang, G. M. Wickramasekara, J. Finney, Smart Grids Are More Efficient: Voltage and Var Optimization Reduces Energy es and Peak s, Vol. 3, ABB Review, 2009. VII. BIOGRAPHIES Xiaoming Feng (SM 2004) is Executive Consulting R&D Engineer with ABB Inc. He has been with ABB Corporate Research for over ten years. He leads research in a broad range of areas in power system simulation, control, and optimization. William Peterson is affiliated with ABB Inc. in Raleigh, North Carolina. Fang Yang (M 2007) joined ABB US Corporate Research Center in Raleigh, North Carolina in 2007. Her research interests include distribution automation, power system reliability analysis, application of artificial intelligence techniques in power system control. Gamini M. Wickramasekara is affiliated with ABB Inc. in Raleigh, North Carolina. John Finney joined ABB in 1995 after receiving a PhD from Georgia Institute of Technology, and has held roles since then as researcher, developer, and research program manager. Currently John serves as product manager for ABB's network management software systems.