Microprocessor -based adaptive water-turbinegovernor

D. Findlay, B.Sc, H. Davie, B.Sc, Ph.D., T.R. Foord, B.Sc, Ph.D., C.Eng., F.I.E.E.,A.G. Marshall, B.Sc, and D.J. Winning, B.Sc, Ph.D.

Indexing terms: Microprocessors, Turbogenerators

Abstract: Conventional governors of water-turbine-generator sets can be set up to provide either stabilitywhen supplying an isolated load or rapid response when connected to a large, predominantly thermal, system.In the paper, an adaptive microprocessor-based governor is described which goes some way to satisfying bothrequirements. Results are given for tests on a 32-5 MW turbine generator.

1 Introduction

The UK electrical Grid system utilises largely thermalgeneration, together with about 4% of conventional andpumped-storage hydro generation. To ensure that suddendemands for electrical power can be met or that unexpec-ted plant loss can be rapidly replaced, it is necessary to havea certain amount of plant connected to the system but onlypartially loaded. The load margin of this plant provides aspinning spare reserve which must be capable of very rapidand sustained increase in output in response to the drop insystem frequency following the demand disturbance.

Conventional hydro and pumped-storage plant have thepotential to be used effectively in this way but suffer onedisadvantage. The governor is often set up so that stabilityis preserved on the infrequent occasions that the generatoris called on to supply load in an isolated part of the systemwhich contains little or no thermal plant. However, whenthe hydro plant is connected to a large, stable thermalsystem, the governor controls the output power of thehydro plant in response to frequency changes in the ther-mal system and has little influence on total system stability.In this role, the hydro governor, with settings which givestable isolated operation, has a somewhat sluggish responseto frequency disturbances and hence does not fulfil its fullpotential for spinning spare.

If, on the other hand, the governor is set up to providevery rapid response to system frequency changes whengrid-connected, the stability of the generating unit onisolated load is lost. The choice of stability or speed ofresponse is one which is made on the basis of systemconsideration and Ashmole et al} suggest that for apumped-storage scheme a compromise should be usedwhich provides system stability when it is run in conjunc-tion with a relatively small thermal system (10 times thecapacity of the pumped-storage plant). In the North ofScotland, however, adverse weather conditions can causesystem 'islanding' in an otherwise secure system, and sothe governor must be set up to permit the sets to supplyisolated loads.

With a view to improving the grid-connected responsewhile maintaining isolated-operation stability, a programme

Paper 972C, first received 8th January and in revised form 18thJune 1980Dr. Foord, Dr. Davie and Dr. Winning are, and Mr. Findlay wasformerly, with the Department of Electronics & Electrical Engin-eering, The University, Glasgow G12 8QQ, Scotland.Mr. Findlay isnow with YARD Ltd., Charing Cross Tower, Glasgow G2 4PP. Mr.Marshall is with the North of Scotland Hydro-Eectric Board

360

0143-7046/80/06360 + $01-50/0

of work has been carried out at the University of Glasgowin collaboration with the North of Scotland Hydro-Electric Board. This resulted in a stable governor2 inwhich the dominant time constant in the power responseto frequency disturbance was reduced substantially. How-ever, even with this substantial improvement, the poten-tially high rate of response of the turbine was still notrealised and work has continued to improve this further andto study the interaction between these improved governorsand a thermal system with a view to the design of governorsfor large pumped-storage installations in the future. As partof this work, a controller testing facility3 has been designedand installed on an operational 32-5 MW water turbine andthis permits site testing of improved governors.

The earlier work has shown that conventional lineargovernors would be inadequate to meet the requirement fora governor which was flexible enough to provide stabilitytogether with a high speed of response. In consequenceit was decided to experiment with governors whose struc-ture and/or parameters varied with changing plant con-ditions. The investigation of these adaptive governors wouldclearly involve complex, changeable hardware, and so thedecision was made to use a microprocessor-based system.This offered the potential, which has subsequently material-ised, of making rapid structural or parameter changes bychanges in program. At the inception of the project, it alsoappeared possible that a microprocessor system mightprovide a cost-effective operational governor; trends inmicroprocessor costs and complexity and the rapid develop-ment of high-level programming languages have whollyconfirmed this prediction.

2 Governor design

In an electrical power network, momentary imbalancebetween demanded electrical power and prime-moverpower will result in acceleration of the generators andhence a change in system frequency. This change, in turn,is detected by the speed governors fitted to all generatingsets and these cause appropriate changes in prime-mcverpowers in an attempt to restore power balance in thesystem. In a water turbine, the governor senses shaft speedor terminal frequency (which is equal to shaft speed if thegenerator remains in synchronism with the system to whichit is connected and ignoring the relatively high-frequencyinter-rotor transients). The governor output acts on thewater control valve through a hydraulic servomotor.

There are many different possible forms for thegovernor. Old designs, many of which are still in oper-

IEEPROC, Vol. 127, Pt. Q No. 6, NOVEMBER 1980

ational use, are entirely mechanical in construction and useflyballs to sense change of frequency. Modern governorsare fully electronic and allow greater adaptability. What-ever physical form the governor takes, its operation is mostreadily understood in terms of its overall transfer function.Thus the very widely used 'temporary-droop' governor maybe represented by the block diagram shown in Fig. 1, forwhich the transfer function is

y_

fl+sTd

(1)

where y is the servo position,/is the frequency error andb , bt, Td and Tyt are the permanent droop, temporarydroop, damping time constant and servo time constantrespectively.4 Governors of this type are characterised bya relatively slow response to a sudden change of frequencybut give stable isolated operation.

fmIf1

— •s i

bt

1

yt

— —

bp

y

Fig. 1 Temporary-droop governorfs = frequency set pointfm = measured frequency

The double-derivative governor used in previous work2'5

is shown in Fig. 2 and includes, in addition to the normalgoverning functions, a servo set-point control and a load-limiting circuit. The transfer function of the governingsection alone, assuming the load-limiting circuit is inactive

(i.e. that yL >^),is

y\ 1f

1 + K, + K2+sT2)

H

[bt-

bP

ybp )(3)

(4)

(2)

where bp and Ty are the permanent droop and servo timeconstant, respectively, Kx and K2 specify the amount offirst and second derivative used and Tx and T2 are filteringtime constants for the derivatives.

In both governors, the response to a step change infrequency is given by an initial, relatively fast, transientfollowed by an exponential rise in servo position, the timeconstant of which dominates the response. The timeconstants of these dominant lags are given by

Temporary droop TL ^

Double derivative TL = T

The dominant lags for the existing mechanical temporary-droop governor at Sloy power station, and the double-derivative governor described in the earlier publication,2

were approximately 160 s and 30 s, respectively; bothgovernors gave stable isolated operation.

A feature which has proved useful in giving addedversatility to the microprocessor based governor describedin Section 3 is that the double-derivatfVe transfer functionreduces to that for the temporary-droop governor whenappropriate parameter values are chosen. For equivalence itis necessary for A"2 — 0. If, in addition, bp is taken to bethe same for both then Tx, Ty and Kx for the double-derivative governor are quadratically related to bt, Td

and Tyt for the temporary-droop governor.In the double-derivative governor shown in Fig. 2, there

are three set points: the frequency set point fs which isused to synchronise the generator, the servo set point ys

which, in a rate-limited form is used in the run-up sequenceand for rapid and precise loading, and the load limiter set

Fig. 2 Double-derivative governor

fs = frequency set pointys = servo set point>>£, = load limiter set point

IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980 361

point j>£, which is also used during the run up and toprovide an artificial ceiling for the servo position for oper-ational purposes.

The load-limiting action is obtained by feeding thedifference between the 'desired servo position' y and theload-limiter set point yL through a one-sided circuit and ahigh gain G to the input of the servo set point. Should yexceed yL, a large signal is fed back and this reduces y x

and hence y with a very short time constant. When yL isgreater than y, the one-sided circuit ensures that no signalis fed back and hence that the load limiter is inactive.

The so-called servo time constant Ty in the double-derivative governor is in fact implemented as part of thegovernor and so the output from the governor, y in Fig. 2,is the desired servo position.

Stabilisation of the governed turbine supplying near tofull power to an isolated load, demands a large value of thedominant time constant TL. However, at lower values ofisolated load, stabilisation is possible with lower values ofTL because the water column in the pipe is now moving ata lower velocity and it is the momentum of this waterwhich dominates the dynamics of the system controlledby the governor. The variation of the optimum controllersettings with load is utilised in the adaptive governordescribed below.

3 Microprocessor governor

3.1 Development

Preliminary studies of the applicability of microprocessorsto the governing of hydro turbines commenced in 1975.Initially a system based on the early Intel 8008 micro-processor was used. Assembly language programs for thisprocessor were crossassembled on a PDP11 computer6 andthe resultant machine code was transferred into and runfrom the read/write memory (RAM) in the microprocessorsystem. Real-time implementations of temporary-droop-like governors were carried out in conjunction with adigital simulation of the remainder of the hydro plant (seeSection 4).

These initial studies confirmed the marginal adequacyof the 8008 processor in implementing simple governorsof the temporary-droop type; however its processing speedwas not sufficient to cope with the more complex transferfunction of the double-derivative type of governor and itsfaster time constants. Around this time a new generation ofmore powerful microprocessors became available fromseveral different manufacturers, and after a comparativestudy it was decided to continue the governor work usingthe Intel 8080 microprocessor. The chief motivationbehind this selection was that the 8080 contained as asubset of its instruction set, the full instruction set of the8008. Hence, all the governor programs for the 8008could, after reassembly, be run on the 8080.

As before, programs for the 8080 microprocessor werewritten in assembly language and crossassembled on aPDP11 computer. The initial testing of these programs wascarried out on an 8080-based general purpose microcom-puter. The machine code output from the crossassemblerwas transferred from the PDP11 and stored on the micro-computer's floppy disc. Thereafter it could be loaded downinto the microcomputer's read/write memory, tested, andquickly amended as necessary to obtain the desired func-tion. In this manner double-derivative governor strategieswere examined and tested against real-time digital plant

simulations with the analogue interface equivalent to thatused on-site.

For all on-site governing of the real hydro turbine, an8080-based Intel SBC 80/10 single-board computer wasused along with a second board containing all of the ana-logue I/O and the real-time clock. Once the governorprogram had been fully tested on the microcomputer,minor adjustments were made to it to enable it to run onthe SBC 80/10 with its different I/O arrangements, and themachine code was written into ultraviolet-erasable pro-grammable read-only memory chips (EPROMs), whichwere transferred to the SBC 80/10. This produced a non-volatile governor program which immediately ran onpowering up the SBC 80/10 system. When site trials showedup possible areas for improvement on the governingstrategies, these EPROMs could be erased and new pro-grams could be written into them.

3.2 Algorithms and software

In the microprocessor governor the differential equations ofgovernors described in Section 2 were solved using thebackward Euler first-order numerical integration method.The short computation time of this simple method permit-ted a rapid updating of the integrated variables, so produc-ing a quasicontinuous output. Backward Euler has also theadvantage that it is numerically stable for all integrationintervals. The derivation of the difference equations for thedouble-derivative governor is given in Appendix 9.

At each fixed increment of time (the integration inter-val) the past and present values of input, intermediate andoutput variables were used in the equations of Appendix9 to compute the new value of the output variable. Thisvalue was transferred to the digital/analogue convertor atthe end of the interval and the procedure was repeated ateach subsequent interval.

All computations in the microprocessor involving thesedifference equations were carried out using real, floating-point arithmetic in preference to integer arithmetic. Thisremoved the need to scale variables and minimised theprobability of numerical overflow. The penalty to be paidfor this approach is the relatively slow speed of floating-point number computations. Typically, floating-pointoperations on the 8080 require several milliseconds and asingle evaluation of the three difference equations describedin Appendix 9 required approximately 50ms. An inte-gration interval of 100 ms was used, thus permitting someother processing computations to take place and stillleave a satisfactory margin.

In order that the difference equations should producean accurate solution of their associated differentialequations, the integration interval T should normally bemuch less than any of the time constants in the differentialequations. The values of Tt and T2 (0-2 s, 0-2 s) used inthe earlier analogue governor clearly did not conform tothis requirement. However, since these are merely the timeconstants of noise-reducing lags which are associated withthe differentiators, their exact value is not critical. Simu-lation studies suggested that factors of two changes inthese constants made little difference to the operation ofthe governor.

Synchronisation of the governor program with real timewas carried out using a crystal-controlled clock whichinterrupted the processor. As a check on the correctsynchronisation of the program, a single bit of digital

362 IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980

output was set at the beginning of each integration intervaland reset at the end of each evaluation of the differenceequations, providing that the time to carry out this evalu-ation did not exceed the integration interval. An externalhardware watchdog timer (see Section 4.6) was attached tothis single bit and if a high-low-high cycle was not completedonce per integration interval, then this fault condition couldinitiate a turbine and generator shut-down sequence.

All the governor types tested were of the general formdescribed in Appendix 9. Load-limiting and servo positionsetting facilities as described in Section 2 were also incor-porated. With a suitable selection of parameters this couldproduce either temporary-droop or double-derivativegovernors and on-site trials were carried out with a range ofprecomputed parameter settings selectable via switches asdescribed in Section 3.3. Table 1 lists the sets of parameterswhich could be employed to produce one of the following:

(a) a temporary-droop governor equivalent to theexisting station mechanical governor

(b) a double-derivative governor equivalent to theelectronic analogue governor of the previous studies

(c) an adaptive double-derivative governor whose oper-ation is outlined below.With the generator operating at low load, a short dominantlag TL in the double-derivative governor can provideisolated load stability and give the desired fast responsewhen grid connected. However, as the load increases,successively larger dominant lags are required, as well asadjustments to the first and second derivative multipliersKx and K2, to ensure isolated load stability. Thus theadaptive governor operates with three sets of parameters,the appropriate set being selected by the measured gener-ator output power. The low-band parameters (see Table1) operated from 0% to 40% of full load, the mid-bandparameters from 40% to 70% and the high-band parametersoperated from 70% to full load (325 MW). The parametersettings in each band were predetermined from computersimulations to provide as short a dominant lag as wascompatible with isolated load stability at the worst oper-ating point in the band (usually the highest load setting).

To prevent instability in the adaptive process when theload is near to the band boundaries, and to further enhancethe speed of response of the adaptive governor, themeasured power signal was fed through a first-order lag —the adaptive time lag — implemented within the micro-processor, before activating the parameter changes. Thus, ifa sharp drop in frequency was observed with the generatorrunning at low load as spinning spare, the load would pickup rapidly with this adaptive lag holding back the tran-sitions to the more sluggish mid and high-band settings. Avalue of 10 s for the adaptive time lag was selected aftersimulations to ensure isolated load stability yet providean enhanced speed of response to grid frequency disturb-ances.

The complete microprocessor governor program includ-

ing the three selectable governor types listed in Table 1along with the necessary floating-point routines was con-tained in 3 kbytes of programmable read-only memory andused a few hundred bytes of read/write memory.

3.3 Hardware

The microprocessor system used for all site trials is shownin Fig. 3. The 80/10 single-board computer was connectedvia the system bus to a second board containing analogueinput/output channels and a crystal-controlled real-timeclock capable of interrupting the c.p.u. The microprocessorgovernor was connected in parallel with an analoguegovernor (see Section 4.6) both being supplied with allinputs at all times. The output of the required governorwas then selected by switch.

The analogue frequency error signal produced by ahardware frequency transducer was transferred to themicroprocessor via one channel of the input analogue/digital convertor while the desired servo position signal wastransferred via one output digital/analogue convertor to theelectrohydraulic servo controller which positioned thewater control valve appropriately (see Section 4.6). Servoset-point and servo limit set-point signals (see Section 2)were generated in both analogue and digital forms in theassociated control equipment (see Section 4.6) and wereread into the microprocessor system via 8-bit paralleldigital input ports.

Additional inputs to the microprocessor but not to theanalogue governor were:

(i) an analogue signal proportional to the electricalpower being generated by the hydro turbine

(ii) the state of user-adjustable switches(iii) the SYNC signal present when the generator circuit

breaker was closed.The power signal was used in the adaptive form of themicroprocessor governor while the switches were used toselect the particular governor type under investigation. Agovernor program for a typical site trial contained severalgoveror types with a range of parameter settings. At eachintegration interval the switches were scanned and theappropriate governor type was selected. This was a furthermajor advantage of the microprocessor governor over theanalogue governors as rapid, bumpless transitions betweengovernor types could be simply achieved. Transitionbetween types could also be made on the basis of the SYNCsignal so that one governor, typically the temporary droop,could be used for run-up control while another was usedsubsequently.4 Plant simulations and on-site test facilities

4.1 General

The hydro plant controlled by the governor system ishighly nonlinear in the following major ways:

(i) The water column dynamics change with water flowand hence with load.

Table 1: Governor parameters for double-derivative structure

Type K, AC, TL = TJbr

1 Temporary-droop

2 Double-derivative

0-40% band

14-99 0 101 0 4-75 003 158-3

3-5 3-5 0-2 0-2 10 0-03 33-3

0-8 2 0 0-2 0-2 0-3 003 100

3 Adaptive 40-70% band < 2-0 3-0 0-2 0-2 0-6 003 200

70-100% band I 3-0 2-8 0-2 0-2 10 003 33-3

IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980 363

(ii) The turbine efficiency is a nonlinear function ofload.

(iii) The water flow is a nonlinear function of the servoposition.

(iv) Backlash exists in the linkage between thehydraulic servomotion and the water control valve (guide-vanes).Thus, although classical control-system design techniquescan be used to give approximate settings, extensive on-sitetests and/or accurate simulation studies are necessary toensure that the governor design takes account of thedepartures in the plant from linear behaviour. To obviatemuch of the time-consuming and expensive site tests and toprovide design tools for future stations, simulations havebeen developed at a number of levels.

In earlier governing studies2'7 a hybrid simulation ofthe plant was used but the problem size and setup andoperating difficulties led to the use of digital simulationsfor most of the work described here. In order to test notonly the philosophy but also the operational hardware ofthe governor prior to control of the real turbine, simulationwas used, not only at the design phase, but also in con-junction with the governor and with progressively greateramounts of plant and equipment on site. These simulationfacilities are now described briefly.

4.2 Offline digital simulation

In this part of the work a full simulation of turbine, pipe-line and servo systems incorporating the nonlinearitiesdescribed above was combined with a simulation of theproposed governing system. The 'Real time interactivesimulation package',12 a Fortran based system developed atGlasgow University for the simulation of continuoussystems, was used for these studies which were performedon PDP11, Prime 400 and GEC 4070 computers at varioustimes. The pipeline system model included a branch to anadjacent generator and used the method of representing thewater column dynamics proposed by Wood.8 The turbine

selectorswi tchesfor governortype

r s ing le board computer

model was that used by Bryce et al1 and included anefficiency characteristic. The servo-system model included anonlinear servo/effective-control-valve-area characteristicand backlash.

This simulation was used interactively, and the governorwas tuned to give stable responses at various load levels.

4.3 Real-time digital simulation

Having designed the structure and chosen the parametersof a governor using the offline simulation, the governor wasimplemented on the microprocessor system, and this wasconnected to a digital simulation on the PDP11 computerof the remainder of the plant. The plant simulationaccepted a 'desired servo position' signal from the governor;the output from the simulation, which was fed to thegovernor, was the frequency error. Since these are theessential interface signals with the real plant, and since thedigital simulation was synchronised to run in real time, themicroprocessor-based site governor was subjected torealistic overall tests in the laboratory without risk toplant.

4.4 Portable plant simulation

Initial tests on-site were conducted using a simple analoguemodel of the plant consisting of the standard simple turbineand penstock models and the rotational inertia equationof the generator together with a linear loss-torque/speedequation which ensured that the servo opening of thismodel at nominal speed matched that of the real plant.

The input to this simulation, see Fig. 4, was derivedinitially from the 'desired servo position' signal from thegovernor and the servo was disconnected. The simulationoutput signal (turbine speed) was fed to a voltage-con-trolled oscillator which in turn fed the frequency trans-ducer in place of the normal signal derived from thegenerator voltage transformer. In this way the governorwas connected to the turbine controller and the remainder

cool re 11 erandpower stati oninterface

p.r.o.m.(A kbytes)

T

r.a.m.(Ikbytes)

c.p.u.(8080A)

TT TT

circuitbreaker

programmableperipheralinterface(A8 lines)

businterface

analogueinput(16 channels12 bits)

analogue

analogueoutput(2 channels,12 b i ts)

clock(i nterrupt)

T T

businterface

-I L

system bus

F i g. 3 Microprocessor system

364 IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980

of the plant, providing an overall check of the governor andits interconnection with station equipment.

Following preliminary checks on system function, theservo was connected with the turbine 'dewatered' and theinput to the simulation was derived from actual servoposition. Using this system, simulated run-up tests wereperformed including all the station equipment except theturbine generator unit itself. Only very occasionally didthese tests fail to reveal problems which subsequentlyappeared with the turbine generator in service.

4.5 Isolation simulation

In earlier work2 the governor stability was checked withthe generator supplying an isolated load. However, the risk

to consumer supply, and the management and cost of suchisolated load tests, militated against their continued use ona routine basis and so the technique of simulated isolationdescribed by Schleif,9 Causon,10 and Brown and Willing11

was adapted for this programme of work.In this technique, with the generator connected to the

Grid, (see Fig. 5) a signal proportional to the electricaloutput power of the generator is fed to the isolationsimulator as an approximation to turbine mechanical-torque output Fm. This is compared, in the simulator, withthe (assumed) electrical torque Fe and the difference, theaccelerating torque Fa, is fed to an integrator with a timeconstant Tm equal to the mechanical time constant of theturbine generator unit. The integrator output is simulatedfrequency fms from which the nominal frequency f0 is

control to/from station

controller

fromgenerator normal

voltage t e s ttransformer

frequencytransducer

microprocessor

governor

,y. (d ig i ta l )

desired

servoposition

t governor/controller system and power station

test equipment

servosystem

servo to

alternatives

t

posit ion turbine

voltage-controlledosci l lator

portable analoguep l a n t s i m u l a t i o n

Fig. 4 Use of portable analogue plant simulation

subtracted leaving the frequency error /. This simulatederror is fed to the governor in place of the normal freq-uency error.

The electrical torque Fe is derived from the total loadpower PL (the sum of the estimated load power PLo and anyinjected disturbance load power APL e.g. test step) usingthe system power-self-regulation factor en in the equation

Fe = ^ (1 + ej)JJ

(5)

generatorinertia I f0(nominal frequency)

powersignalfromgenerator

simulatedfrequency errorto governoi

(test disturbancein load power)

Fig. 5 Full simulation of isolated load

en = load power self-regulation factor

This structure involves one multiplier and one divider, alevel of complexity which is not justified by the use of thesimple approximation of system load/frequency behaviourgiven by the power-self-regulation factor. Using the factthat

fms ~ /o (6)

where fms is the system frequency, / 0 is the set or nominalfrequency (= 1 p.u.) and / is the (small) frequency error,all in per unit, the binomial expansion of l / ( / 0 + / ) gives asimplified expression for Fe of

where

kn = en —

(7)

(8)

Additionally, if APLknf is assumed small compared to, the torque equation further simplifies to

Fe =L0

APL (9)

The system as implemented is shown in Fig. 6. For satis-factory system operation, the switched gain G around theintegrator and the output switch have been added. With

IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980 365

this ganged switch in the position shown, the output isconnected to ground, and the integrator is converted to ashort time constant, first-order lag which thus rapidlyfollows the BALANCE signal, the measure of balancebetween Fm and Fe.

The normal system frequency into the governor ispassed through a dead band of typically ± 0-3 Hz, tempor-arily inserted to ensure that normal system frequencyexcursions do not interfere with the operation of theisolation simulator whilst still affording protection underfault conditions. The simulated frequency error from thesimulator is injected into the governor at the same point asthe normal frequency signal.

The system is put into service by adjusting the gangedpair of PL0 potentiometers (with APL set to zero, thefeedback gain switched in and the output switched off)until Fe equals the value of Fm derived from the turbineand the BALANCE signal is zero. The position of theganged switch is then changed, removing the integratorfeedback and connecting the system output.

Testing on the system is accomplished by injecting loaddisturbance signals to APL and observing simulated freq-uency error and servo position.

Discrepancies between the behaviour of the full-plantsimulation and that of a simulation of the plant with theisolation simulator were found to be because the turbinetorque incorporates a speed-dependent term which isactive in the true isolation case and inactive when isolationis only simulated. Notwithstanding this limitation, thetransient behaviour differed little between the two cases.

4.6 Site-test facilities

The earlier work culminated in tests on a 32-5 MW turbineat Loch Sloy power station on Loch Lomond in Scotland.Since the time of these tests, the temporary test facilitieshave been formalised and a controller testing facility forwater-turbine governors and controllers has been designedand installed.3 In this, an electro-hydraulic servo system hasbeen added to the same mechanical linkage as the originalservomotor and either can be used to operate the watercontrol valve. The controller testing facility also includesall interfaces with the power station control room, auto-matic control scheme and protection, and contains powersupplies, power supply monitors, a 'watchdog timer'monitor for the microprocessor, input/output contactisolation, the servo controller and transducers for frequency,power and reactive power. The system has been designed topermit rapid changeover between the existing stationgovernor and the governor and controller under test

balancemeter

powersignaifromgenerator,- - -

si mutatedfrequency erroito governor

Fig. 6 Simplified simulation of isolated load used in system

in the controller testing facility so that operational use ofthe generating set is not restricted.

Associated with this equipment is a prototype turbinecontroller and an analogue double-derivative governor basedon the design developed in earlier work. The controller isresponsible for generating the appropriate input signalsshown in Fig. 2 for both governors. These input signals:/ = frequency error, ys — servo set point and yL = loadlimiter set point are continuously supplied to the analogueand microprocessor governors which both continuouslygenerate the desired servo position signal y. A switch andbalance indicator on the controller permit the selection ofthe output of one of the governors as the set point for theclosed-loop servomotor controller. This moves the servo-motor to the position demanded by the selected governorusing a position feedback signal derived from the servo-motor.

The frequency error / is the difference between themeasured frequency fm, an analogue signal from the fre-quency transducer, and fs the frequency set point, ananalogue signal generated within the turbine controller.Generation of the set point (or reference) signalsys,yL andfs is the principal function of the turbine controller andinvolves complex logical operations based on digital signalsfrom the power station automatic control equipment, fromoperator controls and from the frequency transducer.

5 Results

5.1 General

A series of simulation studies resulted in governor designswhich were finally subjected to site tests on the waterturbine experimental facility. The results presented herewere all produced during the tests on this 32-5 MW unit.

The monitoring facilities which exist as part of thecontroller testing facility at Sloy include a 12 channel ultra-violet (u.v.) recorder and a 16 channel microprocessor-based data recording system. The u.v. recorder providesimmediate records for analysis and is used in conjunctionwith highly flexible offset and gain amplifiers which permitthe steady component of signals to be offset by a calibratedamount and the remaining signals to be amplified by acontinuously variable and calibrated amount. These recordsalso form a back-up, which, however, has not yet beenrequired, for the data recording system.

The data recording system incorporates a 16 (single-ended) channel, 12 bit accuracy, data acquisition systemand the results are stored on a floppy disk. The softwaredeveloped for this project also permits onsite plotting ofdata on a digital xy-recorder for accurate assessments oftest results. Software usually used offsite includes programsto produce fully annotated results for reports and totransfer the results for comparison purposes to the PDP11minicomputer used for the simulation studies.

5.2 Run-up tests

Since the microprocessor governor was fully integratedwith the experimental controller and with the stationautosequencing equipment, fully autormatic run ups wereperformed with the governor in service. Using test switches,different governor types could be selected for the run up.Fig. 7 shows the run up with the temporary-droopgovernor.

The experimental turbine controller generates the servoset-point signal ys and servo load-limit signal yL (see Fig. 2)

366 I HEP ROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980

and the following sequence of events take place (thenumbers refer to Fig. 7):Before 1: Auto run up has started auxiliaries and opened

main inlet valve.1 — 2: ys and y^ are raised simultaneously from — 5%

to approximately 25% to give breakaway; servoposition y is the sum of ys and the governoroutput which is highly positive but limited to notgreater than yL and so is equal to yL once yL

rises above 0%, the lower limit of servo travel.3: A set speed (approximately 30 Hz) is reached and

servo position is reduced to approximately 21%to provide a better transient response at nearsynchronous speed.

4: Governor action giving negative governor outputcauses >> to be pulled back below yL.

4 — 5: Frequency transient includes start of limitcycling with long period. Frequency reference isadjusted to prepare generator for synchronising.

5 1 - 5 -

event number4 5

time, s

Fig. 7 Run-up sequence

IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980

5: Circuit breaker is closed and controller raiseslimiter7L to 105%.

5—6: Auxiliaries are transferred to unit board.6: Auto run up complete, controller starts loading

set by increasing^ to setting selected during runup.

7: Manual loading of set takes place using control

Run up is performed using the temporary-droop governor,one of the types implemented in the microprocessorsystem, in order to permit easier synchronisation using thepower station automatic synchroniser which is set up forthe normal station temporary-droop governors. The limit-cycle behaviour of frequency is present with all governorsincluding the mechanical type and is due to the backlash inthe control-valve linkages. With the double-derivativegovernor, the amplitude of these limit cycles is smaller butan accompanying shorter period makes automaticsynchronising more difficult. Insertion of a dither signal(1-5% peak-to-peak, 1 Hz) into the servo system has beenfound to eradicate this limit cycling. Injection of dither bythe microprocessor governor when the circuit breaker isopen, is being investigated as an alternative means ofachieving rapid synchronising.

100

100 n

0l_Ltime , s 5C

Fig. 8 Response of governors to 0-97 Hz step drop in frequency,grid connected

a temporary droop (7£, = 160 s)b — -— double derivative (TL = 33 s)c adaptive double derivative

367

5.3 Grid-connected tests

The new governors developed in this work have becomeprogressively faster in response to disturbances in frequencywhen connected to the Grid and the results of a series oftests conducted to demonstrate relative speed of is shownin Fig. 8. The existing station temporary-droop governorshave a dominant time constant of approximately 160 sandcurves a of Fig. 8 show the response of the unit with themicroprocessor equivalent to the station temporary-droopgovernors to a test step reduction of frequency of 0-97 Hz.

The work previously described2 produced a double-derivative governor with a 33 s dominant time constant andcurves b of Fig. 8 show the significant improvement overthe temporary-droop governor. The response rate is limitedinitially by the rate limit on the servo system (set to pro-vide protection for the pipeline) but after about 15%travel at this rate, the servo opening rate becomes limitedby the governor dominant 33 s time constant; the fullpotential of the unit has still not been realised.

The adaptive governor yields the responses given bycurves c of Fig. 8. The servo response is very close to theideal of continuous rate-limited travel. As the powerincreases, the action of the adaptive time constant meansthat the governor remains in each band even after thepower has increased above the top of the band. As a resultthe lower band, and hence faster, governors are used forlonger than the transition of the power signal through thebands would imply.

5.4 System-connected tests using isolation simulator

The adaptive governor was also subjected to a series of testswith the generator synchronised to the Grid but with theisolation simulator in service. The tests were conductedat various load levels and consisted of the injection of 5%step changes in simulated electrical load. kn was set to zerofor these tests, giving a value of en of 1 -0 (see eqn. 8).

When the disturbance power resulted in the total powercrossing the boundary of an adaption band, adaption fromone set of constants to another took place during thetransit. These and all other tests resulted in stable governoroperation and the results of the tests in which adaption-band boundaries were crossed are presented in Figs. 9 and10. In each of these figures a 5% step rise in load is fol-lowed by a 5% step reduction in load and the point ofadaption can be seen in each case.

In Fig. 9, the transition is from the top of the low-band to the bottom of the mid-band, and the low-bandgovernor is clearly only lightly damped at this load level.This however represents the worst possible point from

5% rise —>—change in load power~5°MaU

stability considerations over the whole operating region ofthe governor. At this point, the nonlinear power/servo-stroke curve reaches its maximum gain value, this is thegovernor band with the shortest dominant time lag andhence is the least stable band and the water flow rate isat its highest value within this band. In the continuingprogramme of work on this governor, the possibility isbeing investigated of inversely characterising this non-linear curve within the governor.

6 Conclusions

A microprocessor-based adaptive water turbine governorhas been developed and has been successfully used on a32-5 MW turbine under conditions of run up, Grid-connec-ted operation and operation with a simulation of isolatedload. The use of an adaptive technique permits stableoperation with isolated load together with a rate of powerresponse to system-frequency disturbances limited only byoperational constraints on servo rate.

Governors of this type could be applied to conventionalhydro or to pumped-storage stations. In the UK, their usewith large pumped-storage schmes, especially in areaswhere system 'islanding' is possible, could mean rapidresponse rates but with stability on isolated or near isolatedload conditions.

The use of a microprocessor means that governingstrategies can be readily changed to meet evolving system

5°/o r ise—change in load power—5°/o fall

0 15 r

-1-0

Fig. 9 Governor adaption between low - and mid - bands

368

£ - 1 0 -

Fig. 10 Governor adaption between mid - and high • bands

needs. Although the present governor uses assemblylanguage programs which are relatively complex, newgovernors are under development as part of this project,which use Fortran and a higher level continuous systemsimulation language to define the control action. Theprogram defining the governor will then be written in alanguage which is close to the engineering definition of thegovernor and will also be similar to part of the simulationprogram used to design the governor initially.

Governing is the most complex part of the turbinecontroller which includes run-up control and the operatorinterface. Work is also in hand to integrate these functionswith the governor in a single microprocessor-based system.

7 Acknowledgments

The authors are grateful to the North of Scotland Hydro-Electric Board (NSHEB) for funding and facilities at LochSloy power station where the experimental controllerfacility has been set up.

They are also grateful to the UK Science ResearchCouncil for funding for this work and for a scholarship forMr. D. Findlay.

IEEPROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980

Prof. Lamb is thanked for the provision of laboratoryfacilities at Glasgow University.

The willing and enthusiastic co-operation of NSHEBstaff and of the technical staff at Glasgow University hasbeen of immeasurable benefit in bringing this work to asatisfactory conclusion.

8 References

1 ASHMOLE, P.H., BATTLEBURY, D.R., and BOWDLER,R.K.: 'Power-system model for large frequency disturbances',Proc.IEE, 1974, 121, (7), pp. 601-608

2 BRYCE, G.W., AGNEW, P.W., FOORD, T.R., WINNING, D.J.,and MARSHALL, A.G.: 'On-site investigation of electrohy-draulic governors for the water turbines', ibid., 1977, 124, (2),pp. 147-153

3 WINNING, D.J., MARSHALL, A.G, FINDLAY, D.G.E., AITKEN,K.H., and GRANT, N.F.: 'Controller testing facility on a 32-5 MWwater turbine', IEE Proc. C, Gen., Trans., Distrib., 1980, 127,(6), pp.

4 International Electrotechnical Commission: 'International codefor testing of speed governing systems for hydraulic turbines'.

5 SCHLEIF, F.R., and BATES, C.G.: 'Governing characteristicsfor 820 000 horsepower units for Grand Coulee third powerplant'. IEEE Trans., 1971, PAS-90, pp. 882-888

6 DAVIE, H.: 'Using macroassemblers to create microprocessorcrossassemblers',Microprocessors, 1977, 1, pp. 477-481

7 BRYCE, G.W., FOORD, T.R., MURRAY-SMITH, D.J., andAGNEW, P.W.: 'The use of a hybrid computer simulation inthe investigation of water turbine governors', Simulation, 1976,6, pp. 35-44

8 WOOD, D.J.: 'Waterhammer analysis by analog computer',J.Am.Soc.Civ.Eng., 1967, HYl.pp. 1-11

9 SCHLEIF, F.R., and ANGELL, R.R.: 'Governor tests by simu-lated isolation of hydraulic turbine units', IEEE Trans., 1968,PAS 87, (5), pp. 1263-1269

10 CAUSON, G.J.: 'Governing a hydro-electric system'. Inter-national association for hydraulic research, 7th Internationalsymposium, Vienna, 1974, pp. X/2/1-13

11 BROWN, P.A.N., and WILLING, B.C.: 'Development and useof a machine isolation simulator for testing hydraulic turbinegovernor systems', Inst. Eng. Aust. Elect. Eng. Trans., 1978,EE-14, pp. 20-24

12 DAVIE, H., SCOBIE, D.C.H., and THOMPSON, E.C.: 'AFortran - based simulation package with real - time capabilities'.Proceedings of the United Kingdom Simulation Council confer-ence on computer simulation, May 1975, pp. Al.l—A1.7

9 Appendix: Digital representation of double-derivativegovernor

The differential equations of the double-derivative governorshown in Fig. 2, assuming that load limiting is inactive(i.e. xL = 0), are

Xi = (— xx +Kif)/Ti

-x +^-x \ITX2 + „ x\ 1/ ^2

(10)

(11)

:. +x2)/Ty (12)

The backward Euler method is given by

where T is the integration interval, if1 is the value of theintegrated variable v at the nth integration interval and vn

is the derivative of vn.

Using this method for eqns 10—12 gives

c?-1 +C2{fn-r~l) (13)

c""1 +C 4 (x7-xr 1 ) 04)

'?"! - C 6 ( / " + J C 7 + ^ ) (15)

with

x nx =

c, =

cs =

T+Ti

T+T2

K2

(T +

C6 =bpT

In practice only an approximate solution of the discreteequations above is obtained in real time since the value ofoutput due at time nT can only be output at time nT plusthe computation time for the three equations. This resultsin the introduction of a pure delay, equal to the compu-tation time, which in the presently described governors waseffectively the integration interval T.

IEE PROC, Vol. 127, Pt. C, No. 6, NOVEMBER 1980 369