The purpose of distribution system voltage control is to compensate for load variations and events in the feeding network so that a suitable voltage profile is maintained at every load bus. Many distribution transformers are equipped with a tap changer that can be used to change the effective ratio of the transformer and thereby regulate their secondary-side voltage. Presently, voltage control is based entirely on local criteria, without coordination between different substations. Since the distribution transformers are electrically connected, they interact with each other and as a result, unnecessary operations and oscillations between tap changers at different voltage levels occur frequently. Since each operation causes a voltage step, poor voltage quality as well as wear on the tap changers arises from this interaction.

Several previous studies have shown that the conventional tap changer controller is an important contributor to voltage collapse, since it always aims to restore its controlled voltage, even following severe disturbances in the transmission system. This increases the load on the transmission system and can, as was the case with the Swedish collapse in 1983, contribute significantly to voltage collapse. Since 1993, a licentiate project that resulted in a thesis in June 1997 has been going on at the Department of Industrial Electrical Engineering at Lund Institute of Technology, Lund, together with Sydkraft, Malmo, Sweden. Different methods to improve the control of distribution tap changers have been studied, and a tuning rule for the conventional controllers as well as new centralized control strategies have been derived. A simulation study has shown that the number of tap operations can be reduced by as much as 40% using the centralized control schemes. This was verified in a field test in a distribution branch with three tap changers in the Tomelilla area in the very south of Sweden.

This article outlines the project and highlights some of the academic results of the work. Sydkraft is the second largest electrical power supplier in Sweden, holding approximately 20% of the market. The majority of Sydkraft's electricity is generated at the Barseback and Oskarshamn nuclear power plants, as well as at hydroelectric power stations in both northern and southern Sweden. The Sydkraft Group is an international energy organization with 6100 employees. The utility is split into four business sectors: electricity, gas, heating and services. We produce, distribute and sell electrical power. We also work with natural gas, LPG, solid fuels, data, electrical installations, measuring, telecommunications and consulting services. There are some 19,000 Sydkraft shareholders. The major holders are municipalities in southern Sweden, the German company PreussenElektra and the Norwegian Staatkraft.

The Test System A branch of the distribution system at Tomelilla, a rural area in the south of Sweden, was chosen for the pilot study. Figure 1 shows a schematic of the test system. At Tomelilla (TLA), there is a 100/100/40-MVA, 130/50/20-kV tap-changing transformer. The 50-kV side of the TLA transformer is connected through overhead lines to three 50/20-kV substations, one of which is Jarrestad (JSD). The tap changer in TLA is placed on the 130-kV winding and controls the 20-kV voltage in steps of 1.67%. There is also a capacitor bank present at the 130-kV bus that is controlled based on the time-of-day. At Jarrestad (JED), there is a 16-MVA tap-changing transformer that controls the 20-kV voltage in steps of 2.5%. JSD supplies the Ostra Tommarp (OTP) and five other 20/10-kV transformers mainly through overhead lines. The OTP substation with its 4-MVA tap-changing transformer supplies 42 120/0.4-kV transformers mainly through overhead lines.

Conventional Controllers The typical conventional on-load tap changer (OLTC) controller consists of a time delay and a deadband with hysteresis. The size of the deadband sets the tolerance for long-term voltage deviations and the time delay is intended for noise rejection. Possible tap ratio change is typically 10-15% in steps of 0.6-2.5%. The typical setting of delay time is in the range 30-120 sec whilst the deadband is usually chosen slightly smaller than two tap steps. The mechanical delay time is usually in the range 1-5 sec. The first aim of the project was to find suitable settings of the delay time for the conventional controller. When the project was initiated, the delay times of cascaded tap changers were tuned to the maximum value in Sydkraft's distribution networks, based on the assumption that this would make each OLTC execute as few operations as possible.

Improved Operation The sequence in which tap changers act strongly influences the total number of operations necessary to compensate a given disturbance. The correct strategy is that no OLTC should act until all higher-level OLTCs have compensated for the disturbance. This holds for both load and feeding network step disturbances. Our recommendation for tuning of cascaded OLTCs is: - Set the delay time of the top level OLTC long enough to filter out fast transients. - For a nominal voltage step disturbance at the top level, compute the number of tap operations needed for the top level OLTC to compensate for the disturbance. - For lower level OLTCs, T=NT k + the mechanical delay time. If possible, use constant time characteristic. - Check that the delay time of the lowest level OLTC provides fast enough customer voltage restoration. - Set all but the top level OLTCs to operate in non-sequential mode. The top level may arbitrarily (from the coordination point of view) operate in either sequential or non-sequential mode.

The tuning recommendation yields networks successively longer delay times for lower-level units. Setting the top-level delay time to 30 sec and assuming the mechanical delay time to be a maximum of 10 sec, the tuning rule yields delay times of 70 and 150 sec for the middle and bottom level OLTCs. Figure 2 shows how a 5% decrease in the feeding subtransmission voltage is compensated using the local controllers which all have the standard tuning (every delay time is set to 120 sec). The three OLTCs compensate the disturbance simultaneously, after 120 sec. When the top-level OLTC restores its voltage, it introduces a voltage overshoot at the lower level whose OLTC must make a counteracting operation. Figure 3 shows the response to the same disturbance where the local controllers have been tuned according to the tuning recommendation. With the revised tuning, the voltages are restored quicker, with fewer tap operations and without voltage overshoot at the lower level. Sydkraft AB is now in the process of retuning their controllers according to our recommendation.

Fuzzy-Rule Based Controller We have seen that some degree of coordination can be achieved by retuning the conventional control systems. This, however, is effective only for step disturbances since the time-selectivity assumes that a disturbance makes the voltages go outside the deadband simultaneously for all the OLTCs. For slow ramp disturbances, such as the daily load variations, the voltage varies as a ramp and the lower-level voltages could go outside the deadband before the top level has. Therefore, for effective coordination, a centralized controller using remote measurements is necessary.

Fuzzy control has been most successfully used for control problems where the control objectives are difficult to quantify or where one has some heuristic knowledge that can improve control. The basic idea is that an observation of each physical process variable can be translated to a fuzzy variable, giving it a linguistic interpretation. Using fuzzy inference and approximate reasoning, the controller outputs can be determined on the basis of a number of measurements and a set of heuristic rules.

Using observations from simulations with the conventional controllers, the following heuristic rules that express some desired properties of the controller were derived: 1. If the (local) voltage is high, order a downward tap operation. 2. If the (local) voltage is low, order an upward tap operation. 3. Cancel an upward tap operation if any tap changer higher up in the network is about to order an upward operation. 4. Cancel a downward tap operation if any tap changer higher up in the network is about to order a downward operation. 5. If voltage deviation is very large, order an operation regardless of Rules 3-4.

Note that Rules 3-4 introduce the need for a centralized controller based on remote control and measurements.

Developed Controller Prototype To validate the simulation results and to demonstrate the implementation of the coordinated control schemes, a prototype controller was developed and installed in the test system at Tomelilla. It is a distributed control system, with a local control loop in each substation. A supervisory controller in the Tomelilla substation coordinates the local control loops with the necessary communication between the substations provided by radio links. Each local control loop can operate autonomously using a conventional (local) control system whenever contact with the supervisory controller is lost. The supervisory controller was implemented using Wonderware's InTouch, a rapid development tool for construction for SCADA systems. The local substation controllers were implemented using LonWorks components. The sampling time of the controller is one minute.

Measurement And Simulation Using the data recorded in the field test, the different control schemes can be compared in terms of the number of tap operations required and the resulting voltage deviations. Figure 4 shows the daily average number of tap operations recorded in the simulations and field measurements during a two-week measurement taken in the autumn 1996. During the first week, the conventional controllers with the revised tuning were used and during the second week, the fuzzy rule-based controller was used.

>From the figure we see that retuning the local controllers yields a reduction of about 11% compared to the same controllers with the conventional tuning. The corresponding reduction for the fuzzy rule-based controller is about 38 % (30% compared to conventional controllers with normal tuning). The measurements indicate roughly a 40% reduction compared to the retuned conventional control. The greater reduction observed in the measurement than in the simulations is due to the fact that the measurement with the fuzzy-rule based controller was made during the second week when all the controllers needed slightly fewer operations than during the first week. The simulations are all based on data recorded during both weeks, and are therefore more reliable. We can also see that the major savings are done at the lowest level where the number of units is the largest. In the experiments, the different control schemes were tuned to give the same voltage deviations from setpoints (in terms of voltage standard deviation). Figures 5 and 6 show the voltages and tap positions during one day of the measurement period.

Conclusion In the project, simulation and measurement results have shown that the control of cascaded tap changers needs to be coordinated in order to avoid tap changers counteracting each other. Failure to provide this coordination will cause over- and under-voltages as well as increased wear on the tap changers. It has been shown that some coordination can be obtained by proper tuning of the existing controllers. A two-week field measurement where the conventional controllers were compared to a new centralized control scheme, has shown that the average daily number of tap operations is reduced by about 40% compared to the existing controllers

Mats Larsson is a post graduate student at Lund Institute of Technology and Power Technology, Malmo, Sweden where he received his masters degree in computer science and engineering and is with WM-Data Ellips AB. He is a technical licentiate and is with Lund's Department of Industrial Electrical Engineering and Automation. His licentiate project has been on analysis and practical implementation of coordinated tap changer control schemes for distribution networks. He is active in the department's Distribution Automation project.

Daniel Karlsson is chief engineer, system operation at the Power System Analysis Group with the Operation Dept. of Sydkraft, the largest private power utility in Sweden. He received his master's degree and licentiate degree and the Ph.D. degree in electrical engineering, from Chalmers University of Technology, Sweden. His work has been in the protection and power system analysis area and the research has been on voltage stability and collapse phenomena with emphasis on the influence of loads, on load tap changers and generation reactive power limitations.