00493519 Interconnected System Frequency Response

Interconnected System Frequency Response Rogelio Luck Department of Mechanical Engineering Mississippi State University Mississippi State, MS 39762-9550

Roger L. King and Minh-Luan Ngo Department of Electrical & Computer Engineering Mississippi State University Mississippi State, MS 39762-9571

Abstract Frequency management is a primary goal in the control of an electric power system. Traditionally, this has been accomplished via tie-line bias control within a well defined control area. However, the control area 's frequency bias characteristic, which is an important component in the tieline bias control equation, is a dgficult term to quantih. This is due to the fact that this term represents the combined droop characteristics of all the generators serving the load plus the frequency dependency of the load. However, since most load frequency controllers use integral controllers, this inability to accurately quanti& the control area 'sfrequency response characteristic has not caused unacceptable control errors. But, as the power industry moves into a new era, the control schemes must be re-examined in the light of the new system operating rules. This paper will introduce some new ideas on frequency management that utilize what is referred to in this paper as the interconnected system frequency response characteristic, ,(jsy9 This paper will explain the problems associated with the present approach and why this new method offers improvements for large control areas, especially in the new competitive environment. An analytical approach for determining ,(jsclsys in real-time will be shown. This approach has been developed with system load data from the Southern Electric System. Keywords: Frequency bias coefficient, Area generation control, Area control error

Control areas and AGC The electric utility system in the United States is divided into many interconnected control areas. Each of these areas is responsible for generating enough power to meet its own customers or "native load." By keeping the generated power equal to the power consumed by the load, utilities keep the overall system frequency at 60 Hz. Under-generation will cause the frequency to drop below 60 Hz while over-generation will cause the frequency to rise above 60 Hz. Since the loads in the system are constantly changing, the

generation level in each area must be changed as well, in order to maintain the desired frequency. The various control areas are interconnected by transmission lines called tie-lines. Periodically, the different areas contract sales of bulk power from one to another across these tie-lines. For example, one area may agree to buy 100 MWs of power from its neighbor for the next hour. The two areas then adjust their generation levels in order to cause the 100 MWs to flow out of the selling area and into the buying area across the tie-line. This practice adds another factor to generation control. Not only must areas adjust their generation to meet their own changing native load, but they must also maintain any scheduled tie-line transactions. It is possible, by monitoring both the tie-line flow and the system frequency to determine the proper generation action (raise or lower). Thus, electric utilities use an automatic generation control (AGC) system to balance their moment-to-moment electrical generation to load within a given control area. The primary objectives of AGC are: (a) to keep tie line interchanges as scheduled, (b) to maintain the system frequency at the nominal frequency (60 Hz in the U.S.A.), (c) to operate with security in mind (i.e., with sufficient reserve to accommodate disturbances in demand or generation), (d) to optimize economical operation, and (e) to maximize compliance with North American Electric Reliability Council (NERC) guidelines [ 11. Regardless of these guidelines, any good controller for load-frequency balance should be designed to reject frequency and load disturbances, i.e., items (a), (b), and (c). NERC guidelines exist to enforce items (a), (b), and (c) in a manner consistent throughout the interconnected system by establishing bounds in the performance of the AGCs.

Tie-line bias control The current practice of the load frequency control (LFC) h c t i o n of automatic generation control (AGC) is based on a strategy known as tie-line bias control. In this control strategy each area of an interconnected system tries to regulate its area control error (ACE) to zero, where:

306 0-8186-7352-4/96 $05.00 0 1996 IEEE

Then, since a portion of the load is frequency dependent, the final load will be less than that at rated frequency. Since a control area's p is continuously changing and no The term T,-T, is the difference between the actual and the good technique has been developed to measure it in real-time, schleduled net interchange on the tie lines. The term it must be estimated. The estimated value of p is called the representing the area's natural response to frequency Tie-Line Frequency Bias Coefficient, B [3]. The closer B deviations is lOp(f,-f,). The coefficient, p, is known as the matches p, the better AGC will be able to reduce the number system natural response coefficient. It is difficult to obtain of unnecessary control actions and minimize inadvertent an accurate value of p since it depends on the governor energy flows. NERC guidelines require that the annual reslponse capability of the generating units presently on-line estimate for B be based on the average of the area's B, as and the frequency dependence of the constantly changing observed for past disturbances during peak hours. NERC load. This characteristic is expressed as: also requires that the monthly average of B should not be less then 1% of the control area's annual estimated peak load p = -1 + D [I]. Therefore, it is common practice for a utility to use 1% R of its annual peak load as its value for B in the ACE equation (equation 1). where, The generator's droop characteristic and the frequency 1iR is the generator regulation or droop, dependency of the load permit frequency management of a control area to be somewhat self-regulating. However, since D is the load damping Characteristic. there is a strong desire to keep frequency at its nominal R is the steady state characteristic relating frequency of a value, there must be a supplemental control system (AGC) to generator to generated power (see figure 1). Ideally, the adjust a unit's output to bring the steady state frequency back system individual unit droop characteristics are set to be to nominal. Since the control areas operate as an interconnected about equal to minimize turbine governor oscillation. Typically, droops are designed to range from 3 to 5 percent. system, it is important for a particular control area's AGC to In other words, a governor droop characteristic of 5% means be able to determine the source of a disturbance (i.e., inside a 5% change in frequency (e.g., 3 Hz on a 60 Hz system) or outside of its area). This task is accomplished with the will result in a unit cycling its output from zero to full load. ACE equation and tie-line bias control. Using the system However, according to a recent EPRI survey, the droop frequency and the net power flowing over the tie-lines, AGC characteristics of the generators measured in the survey can differentiate between an internal and external disturbance. actually ranged from 6.9 to 12.2 percent [2]. These values If it is an internal disturbance, it will be the task of the local are much higher than the commonly anticipated range of 3 to AGC to adjust area generation to meet the load. If it is an external disturbance, the control area's obligation is to 5 percent. provide power to the affected control area in a proportion equivalent to the control area's natural system response coefficient, p.

1 I

r

I

4

I

Actual vs estimated ACE

U

P

0

Unit output

Figure 1 Speed droop characteristic of a governor.

I) represents the self-regulating characteristic of the load. It is expressed as the percent change in the connected load divided by the percent change in frequency. In-other-words, as system load increases the system frequency decreases.

The actual error between generation and load of an area is the area's actual ACE. The estimated ACE is calculated using the estimated values of the tie line interchange, frequency deviations from the nominal, and the estimated frequency bias factor, B. It is this estimated ACE that is used as a control variable by AGC. When the frequency bias of the estimated ACE matches with the actual frequency bias of the area, the estimated ACE and the actual ACE are identical [3]. Since one of the objectives of AGC is to regulate generation, its input should be the actual ACE. It can be argued that integral action in AGC results in both actual and estimated ACE being averaged to zero over time, but the transient response of AGC can be significantly degraded by errors in the frequency bias factor. For example, variations in the estimated ACE can be much larger

307

than variations in the actual ACE, if the frequency bias term in the estimated ACE is too large. Also, an estimated ACE based on a constant frequency bias factor is non-linearly related to the required change in generation because the actual frequency bias factor is non-linearly related to the operating state of the area, i.e., frequency and load. Using an estimated ACE based on a constant frequency bias setting as the system error in AGC introduces an undesirable nonlinearity into the control system. To illustrate these points, the estimated frequency responses, 10B(f, - fs), for the Southern Electric System (SES) were calculated for the month of July 1994. For 1994, SES used a constant B coefficient of 321 MW/O.lHz. The calculated values are the average frequency responses over a thirty days period. The calculated values of the frequency response can be seen in Table 1. Frequency Response Term of ACE

I

1 E; I 11

Standard Deviation (MW)

SES'SB

I

I

+lo% B

I

-0.4909

28.7729

I

I

-10% B

I

-0.5399

3 1.6509

-0.4418

1 II

25.8961

Table 1: Calculated Frequency Response with Different B's for the SES July 1994 data. Table 1 shows that the mean value of the frequency response term of the ACE equation actually used on the SES is less than 0.5 MW. This is as expected since integral action of the controller should work to zero out the error. Note that 95% of the excursions for the month were within 57.6 MW (i.e~,two standard deviations). However, note that if the B coefficient was 10 percent higher (and NERC prefers utilities to err towards supplying additional frequency response), then the mean would worsen by 10 percent. In this case, two standard deviations would be 63.3 MW. Conversely, if the B coefficient was 10 percent less than what is presently used, an improvement in the inadvertent energy interchange and an improvement in the magnitude of the excursions would be noted.

Also, since control areas are interconnected, it can be shown that the change in the system steady state frequency due to an incremental change in load power in a particular control area is:

Am

=

- APLI

P,

+

P*

+

a * *

+

P"

(4)

The assumption that the system frequency is constant is valid if the tie-lines interconnecting the control areas are stiff. To illustrate the frequency support offered by operating in an interconnected power system consider the following example (see figure 2). First, consider the case where Area 1 has its three tie-lines open and is operating in isolation and there is no AGC operating. Assume this control area's natural response characteristic, p, is 300 MW/O.l Hz based upon the 1% of peak load guideline. Now, what happens if the area experiences a 10 MW increase in load? From equation 3, it can be seen that the steady state frequency deviation will be a -0.00333 Hz. The decrease in frequency is expected due to the fact no AGC was available to change the generator outputs to match the 10 MW load increase. Now consider what happens if this same system is interconnected with three smaller control areas with p's of 75 MW/O.lHz and Area 1 again experiences a 10 MW increase in load. Again, there are no AGCs operating in any of the control areas. From equation 4, it can be determined that the interconnected system has a steady state frequency deviation of -0.0019 Hz. This is a 43.4% improvement in frequency management due to the interconnection. Area 3

Frequency support In textbooks dealing with the subject of automatic generation control [4] it is shown that the change in steady state frequency due to an incremental change in load power in a particular control area is:

Area 2

Area 4

Figure 2 Four control areas interconnected. However, this frequency support does not come for free on the interconnected system. Using the same scenarios postulated above, it can be seen in the isolated control system that for a 0.00333 Hz chnage in frequecny, the units

308

on-line in Area 1 will increase their outputs by 9.99 MW due technique used past load data (past several minutes, In the yesterday, and the same day last week) and the AGC time to droop and load frequecny dependency. interconnected scenario, where the frequency deviation is cycle (every 6 seconds) to predict the load change that was 0.0019 Hz, the units in Area 1 will increase their outputs by dependent upon time. This can be thought of as: a total of 5.7 MWs, while the three smaller areas are 6L required to increase their outputs by 1.425 MWs each. The &=---At 8t total increase in generation in this case is 9.975 MWs for the 10 MW load increase. Therefore, Area 1 only had to provide 57% of the generation for the short term frequency support instead of all of it. The other three areas provide the where, balance of the power for a problem that is not theirs. AL is the actual change in load, In the interconnected case, the larger system (Area 1) did 6L/6t is the time dependency of the load, noi have to provide all of it's own frequency support due to At is the time between each AGC cycle. the fact it could lean on the ties. In fact, it experienced a 42.9%decrease in the amount of its own generation used for With the success of this approach, it became apparent that a frequency support. However, it needs to be realized that this similar technique might be feasible for predicting p. leaning on the ties for frequency support is only a short term Since a load disturbance results in a frequency deviation, solution. The AGC of Area 1 needs to recognize the it was postulated that the change in load had not only a time shortfall in generation and adjust generator outputs to pick up dependent component, but also a frequency dependent the 10 MW load increase and bring the system frequency portion. This can be thought of as: back to its nominal value. However, it also needs to be recognized that if any of the AL = -6L At + 6L AO 6t 60 three smaller areas had a similar 10 MW increase in load that Area 1 would supply 5.7 MWs towards the system frequency support. Never-the-less, under the old operating philosophy, all of these inadvertent energy exchanges were accounted for where, and the control area that needed to lean on the ties paid the 6 U 6 o is the frequency dependency of the load, grid back in-kind at an appropriate time. A o is the change in frequency between each AGC cycle. A complication arises now that the Energy Policy Act of 1992 has opened the power generation market to competition. Examining equations 3 and 6 shows that the 6L/6o term Now, it must be determined how to handle frequency in equation 6 is equivalent to the system natural response management with more units coming on-line as independent coefficient, p. Note also, that the denominator of equation power producers (IPPs). For power system operating and 4 dictates that the interconnected system frequency deviation control purposes it is possible to handle IPPs as just another is a function of all of the droop characteristics of the units tie-line. In-other-words, there should be a certain amount of on-line and the overall system load damping characteristic (see equation 2). So, when making an attempt to predict a power scheduled for that line. Looking at figure 2 again, what happens if one of the particular control area's system natural response coefficient, areas is now just a generating unit? Assume a frequency p, from its load changes (APJ and system frequency excursion occurs and an area leans on the ties for short term deviations (Am) the value actually predicted is what is frelquency support. How will the control area who was the referred to in this paper as the interconnected system recipient of frequency support from an IPP pay energy back frequency response characteristic, psys and it will be defined to ihe IPP? Since an IPP is only an energy supplier, it would as: appear that it would seek a monetary payback for the (7) P*s = P, + p* + ... + p" inadvertent energy. Therefore, as utilities price their ancillary services, this new complication to the operation and control of the bulk power system must be considered. Therefore, equation 6 can be simplified and rewritten as: Estimation of ssYs

AL

Previous research [5] had shown that it was feasible to make accurate very short term (-10 minute) load forecasts froim actual load data. The technique utilized a p-step ahead autoregressive prediction with a least means squares recursive algorithm and an exponential forgetting factor. This

=

AA? + , p

AO

(8)

When actual past AGC data is used, equation 8 can be rewritten in matrix form as:

309

AL

At -

ALK

Hz of support and the other two areas pick up 112.5 MWIO. 1 Hz each. Another scenario, is that there may be IPPs wanting to sell frequency support (e.g., an IPP may offer to pickup 500 MW/O.1 Hz of frequency support). Never-theless, whatever is decided there needs to be a method to verify the interconnect has proper frequency support. This technique of estimating psysmay be that method.

Aa

AtK AmK

[p”,]

(9)

where, AL is the present load change, AL, is the load change at the Kth past AGC cycle, At is the change in time between data points. If all the AGC data is used, then this is a constant equal to the AGC cycle time (e.g., 6 seconds)., Am is the frequency deviation associated with the present load change, A% is the frequency deviation associated with the oldest load change. As stated previously, a separate algorithm has already been developed for the very short term load forecasting portion of equation 9. In other words, the methodology for finding A has already been developed and is different from that described herein for psYs. One reason for this is that Aa typically has to be measured over several AGC cycles. This requirement is due to the fact there is very little change in frequency from one AGC cycle to the next cycle. In fact, this change is so small that it initially gave divide by zero errors in the prediction algorithm. Equation 9 can now be solved by a least squares estimation technique with the resulting value for psys being in units of MW/Hz [6]. It was found that at least 30 minutes of past data were required in the calculation to obtain Rz correlations of significance. Values for psys varied significantly based upon the time of day the calculations were made. However, it is important to note that these numbers were normally less than the B used by the Southem Electric System in the ACE equation. The reader is reminded to review the findings demonstrated in table 1 about how the magnitude of B contributed to the swings in the frequency term of the ACE equation. Normally, large swings in ACE are accompanied by control actions. The reason for calculating psysis that since this value represents the interconnected system’s frequency response characteristic (see equation 4), then this value may be important in determining how much frequency support a given control area should be required to supply to the interconnection. For example, psys in the example of figure 2 is 525 MWlO.1 Hz according to the NERC 1% of peak requirement. In the new deregulated environment, frequency support might be negotiated between the four control areas in a different manner (e.g., 13 1.25 M W / O . 1 Hz each) or if an IPP is involved they may not want to participate in frequency support at all. In this case, Area 1 might take 300 MWlO.1

P b q W ,xbq, Frequency management of the interconnected power system will need to be maintained even as the electric utility industry undergoes significant changes in its power generation options. This service will probably be bought or sold by energy suppliers and transmission companies as an ancillary cost of doing business. It will be important for all participants to understand the importance of frequency management and how it works on an interconnected system. It will also be important for companies that buy or sell these service to confirm that they are getting or have available to them the services for which they have paid. The interconnected system frequency response characteristic, psys, is a possible solution.

The research reported in this paper has been supported by grant number ECS-92-16549 from the National Science Foundation and contract number RP 8030-12 from the Electric Power Research Institute. The data for the ACE calculations was supplied by Southem Company Services.

References 1. North American Electric Reliability Council Control Performance Criteria Training Document, May 3, 1991. 2. “Impacts of Governor Response Changes on the Security of North American Interconnections,” EPRI Final Report October 1992. 3. N. Jaleeli, D.N. Ewart, and L.H. Fink, “Understanding Automatic Generation Control, “IEEE Transactions on Power System, Vol. 7, No. 3 August 1992, pp. 11061122. 4. A.J. Wood and B.F. Wollenberg, Power Generation, Operation, & Control, John Wiley & Sons, 1984. 5. R.L. King and R. Luck, “Intelligent Control Concepts for Automatic Generation Control of Power Systems,” NSF Annual Report ECS-92-16549, March 31, 1995. 6. Minh-Luan D. Ngo, “Determination of the Interconnected System Frequency Response Characteristic,” MS Thesis, Mississippi State University, December 1995.

310