Neuroidentification of system parameters of the UPFC

Neuroidentification of System Parameters of the UPFC in a Multimachine Power System Radha P. Kalyani Department of Electrical and Computer Engineering University of Missouri - Rolla, MO 65409, USA rpk5p@umr.edu Abstract The crucial factor affecting the modern power systems to- day is loadflow control. The Unified Power Flow Control- ler is an effective means for controlling the power flow. The UPFC is controlled conventionally using PI Controllers. This paper presents the designs of neuroidentiJiers that models the system dynamics one-time step ahead making the pathway for the design of adaptive neurocontrollers. Two neuroidentiJiers are used for identifjling the nonlinear dynamics of a multimachine power system and UPFC, one neuroidentifer for the shunt inverter and another for the series inverter. Simulation results carried out in the PSCAD/EMTDC environments on multimachine power sys- tem are presented to show the successful neuroidentifcation of system dynamics. Keywords Multimachine Power System, Unified Power Flow Control- ler (UPFC), Neuroidentification, Scries Neuroidentifier, Shunt Neuroidentifier, Adaptive Control. INTRODUCTION With the ever-increasing complexities in power systems across the globe and the growing need to provide stable, secure, controlled, economic, and high-quality electric power -especially in today’s deregulated environment - it is envisaged that Flexible AC Transmission System (FACTS) controllers are going to play a critical role in power systems [I]. FACTS enhance the stability of the power system both with its fast control characteristics and with its continuous compensating capability. The two main objectives of FACTS technology are to control power flow and increase the transmission capacity over an existing transmission cor- ridor [2]. Gyugyi proposed the Unified Power Flow Controller which is a new generation of FACTS devices in 1991 [3]. It is a device, which can control simultaneously all three parame- ters of power transmission line (impedance, voltage and phase angle). This device combines together the features of two other FACTS devices: the Static Synchronous Compen- sator (STATCOM) and the Static Synchronous Series Com- pensator (SSSC). Practically, these two devices are two Voltage Source Inverters (VSI’s) connected respectively in shunt with the transmission line through a shunt transformer and in series with the transmission line through a series 0-7803-8243-91041$17.00 0 2004 IEEE 243 Ganesh K. Venayagamoorthy Department of Electrical and Computer Engineering University of Missouri- Rolla, MO 65409, USA gkumar@ieee. org transformer. These are connected to each other by a com- mon DC link, which is a typical storage capacitor. The shunt inverter is used for voltage regulation at the point of connection, injecting reactive power flow into the line and to balance the real power flow exchanged between the series inverter and the transmission line. Thereby, the UPFC can fulfill functions of reactive shunt compensation, active and reactive series compensation and phase shifting. Be- sides, the UPFC allows a secondary but important function such as stability control to suppress power system oscilla- tions improving the transient stability of power system [2]. The ability to learn and store information about a physical plant allows neural networks to be used in modeling and designing power system controllers [4, 51. These offer an alternative to conventional controllers. Neural networks are suitable for multi variable applications as they can easily identify the interactions between the system’s inputs and outputs. The application of neural networks in power sys- tems arises due its inherently good property of pattern rec- ognition, rapid performance of multiple-input multiple- output calculations and the ability to learn and store infor- mation about physical plant. This paper presents the design of two neuroidentifiers which can be used in the further design of neurocontrollers to replace the conventional PI controllers in the shunt and series branches of UPFC. The advantage of neurocontrollers over conventional controllers is that for changes in the op- erating points and system parameters, the neurocontrollers can adapt their parameters accordingly to the system changes automatically unlike conventional controllers which require human intervention. MULTIMACHINE POWER SYSTEM For studying the control of a UPFC in a multimachine power system, the setup shown in Figure 1 is simulated in the PSCADEMTDC environment. The power system in Figure 1 consists of two synchronous generators GI and GZ of ratings P1=l 600MVA and P2=2200MVA respectively along with exciter and govemor models connected to an infinite bus and two loads, one of value P (real power) =3000 MW, Q (reactive power) = 1800 MVAR and another of value P = 3000 M W , Q = 300 W A R are connected at bus 2. IClSlP 2004 X=0.2 on 5000 MVA 5000 MVA base 3566 MW & 3-t Infinite Bus 500 kV \ Bus, 5000MW 500 kv 500 kV Bus, - Bus, Bus, 2094 MW Industrial Load 3000 MW 1800 MVAr T U 1 ' 1 ' ' -1500 MVAr 0.03+j0.1 on T I 3300 MVA base - p. .! kv Bus, BUS, Bus, I ., Residential and Commercial load 3000 MW MVAr ... - . . Load area Figure 1. Multimachine power system with UPFC installed between buses 2 and 3. UNIFIED POWER FLOW CONTROLLER UPFC is a generalized synchronous voltage source (SVS), represented at the fundamental frequency by voltage phasor V with controllable magnitude V (05 V 5 Vmax) and angle a (05 a 5 2n), in series with the transmission line. The UPFC consists af two voltage-sourced inverters. These back-to-back inverters are operated from a common DC link provided by a DC storage capacitor. This ar- rangement functions as an ideal ac-to-ac power inverter in which the real power can freely flow in either direction between the ac terminals of the two inverters, and each inverter can independently generate (or absorb) reactive power at its own ac output terminal. The series inverter provides the main function of the UPFC by injecting a voltage V with controllable magnitude V and phase angle a in series with the line via an insertion trans- former. This injected voltage acts essentially as a synchro- nous ac voltage source. The transmission line current flows through this voltage source resulting in reactive and active power exchange between it and ac system. The inverter generates the reactive power exchanged at the ac terminal internally. The active power exchanged at the ac terminal is converted into dc power, which appears at the DC link as a positive or negative real power demand. The basic function of shunt inverter is to supply or absorb the real power de- manded by series inverter at the common DC link to sup- port the real power exchange resulting from series voltage injection. This DC link demand of series inverter is con- verted back to ac by shunt inverter and coupled to the transmission line bus via a shunt-connected transformer. In addition to this the shunt inverter can also generate or ab- sorb controllable reactive power, if it is desired and thereby provides independent shunt reactive compensation for the line. The single machine infinite bus power system along with UPFC is as shown in the Figure 1 [3]. The three main control parameters of UPFC are magnitude (V), angle (a) and shunt reactive current control of real and reactive power can be achieveld by injecting series voltage with ap- propriate magnitude and angle. This injected voltage is transformed into dq reference frame, which is split into Ed and Eq. These coordinates can be used to control the power flow. Following sections of the paper discusses the control struc- tures of shunt and series branches according to which the identification of controlling parameters is carried out. Shunt Branch Control Control of the shunt inverter is achieved by varying the . shunt inverter voltage active and reactive components Epd and Epq appropriately. The dq reference transient stability model for UPFC shunt input circuit is as follows: L J (1 1 Assuming that R,1 <<: L,I we can wiite the above equation in steady state: 244 IClSlP 2004 Hence it can be observed that reactive power s u p p j and shunt input voltage can be regulated by active voltage com- ponent Epd and the DC-link capacitor voltage support can be achieved by regulating Epq. Shunt Inverter I PWM I Figure 2. Shunt branch of UPFC with PI controllers (SI and S2 at position 1) and PRBS signals (SI and S2 at position 2). Series Branch Control The control of series inverter can be achieved by PQ de- coupled control. Neglecting invcrter losses, the injected active power and reactive power as well as output powers are given by PQ decoupled control. Neglecting inverter losses, V(Eq -Eq cosS+ E, sin@ x p,' ( 3 4 VE, cos Si- VEq sin S - VE, + Ed2 + Eq2 x Q, = V2sinS+VEq x e,, = 2VE, cos S i- 2VEq sin S + E,' i- Eq2 2x Qou, = where V, =JE,7+E,z E, = V, sin(8,) E~ = v, cos(e,) Equation (4a) shows that Po,, is mainly affected by Eq whereas (4b) shows that QOut affected by both Eq and Ed. In incremental form, the line active and reactive power can be expressed in terms of AEd and AEq. V out x AP =-Mq ( 5 4 1 Wout=$AEd Vcos G+AEqVsin8+AEdEdo +M,E,,) (5b) (6) 1 AQout=F(md v+md Edo +mq 'qo) The control of the active and reactive power on the trans- mission line can be achieved using the decoupled algo- rithm. The block diagram of PQ decoupled series inverter controller is as shown follows [6]. ri Series Inverter Figure 3. Series branch of UPFC with PI controllers (SI and S2 at position 1) and PRBS signals (SI and S2 at position 2). DESIGN OF NEURO-IDENTIFIERS The neural network architecture for the control of UPFC consists of two identifiers one for the shunt inverter and another for the series inverter. These networks dynamically identify the controlling parameters of UPFC, AEd and AEq which are the outputs of PI controllers. The Neuro- identifier is developed using the series-parallel Nonlinear Auto Regressive Moving Average (NARMA) model [4]. The output of this model is 5 at time (ki-1) depends on both past n values of output and m past values of input. The neuroidentifier output equation takes the form given by y(k),y(k -l), . . . . . . .,y(k -TZ + 1) u(k),u(k - l), . . . . . . .-. ,u(k - m+ 1) ?(k + 1) = here y(k) and u(k) represent the output and input of the plant to be controlled at time k, respectively. The NARMA model is used in preference to other system models because online training is desired to correctly identify the dynamics of the UPFC and therefore avoiding a feedback loop in the 245 IClSlP 2004 model, which allows static back propagation to be used to adjust the neural network weights. This reduces the compu- tational overhead substantially for online training [4]. Shunt Neuroidentifier The shunt UPFC branch neuroidentifier (SHNI) in Figure 4 is a three layer feedforward neural network with thirteen inputs, a single hidden layer with fifteen sigmoidal neurons and two outputs. There are four different types of inputs, the first two inputs to the NI are namely, the deviation sig- nals between the measured shunt voltage and its reference value Verr, the measured dc link voltage and its reference Vdcerr, and the other two types are the PRBS training sig- nals AEpdgrbs 'and AEpqqrbs (Figures 5 and 6 ) with magnitudes in proportion to the real and reactive compo- nents of shunt inverter voltage Epd and Epq respectively. SHNI ; ::'Shunt , Neuroidentifier AE pd- prbs AEpq-prbs vdcerr(f) PLANT D t m c t ) , E vdcen@) I .................. hill ! I i Figure 4. Neuro-identifier for shunt branch. . ... . . . . . . . . . . . . . . . . . +o,8i- -7 -0.64- ._ . 11.4 -12.44 13.48 14.52 7556- 16.6 Time (sec) Figure 5. PRBS signal Epd applied to shunt branch of UPFC. ..... .. . ... ........ , , I / / I / ........... 13:32 ..... ...... 11.4 12.36 14.28 15.24 16.2 Time (sec) Figure 6. PRBS signal E,, applied to shunt branch of UPFC. All the fdur types of inputs are time delayed by one sample period and together with their eight previously delayed values form the twelve inputs to the SHNI at C (Figure. 4). The outputs of the SHNI at E are the shunt voltage devia- tion 'em and dc link voltage deviation 'deem which are estimated one time step ahead. These PRBS signals are only fed to the shunt inverter at C and plant at B at during the pre-training phase with the aid of switches S1 and S2 (Figure 2). The outputs of SHNI at E are compared to out- puts of the plant at D and the error signals at F are used to update the weights of' the SHNI using the backpropagation algorithm. This process is repeated until a satisfactory error goal is obtained with the SHNI learning over a number of different possible operating points of the plant. Series Neuroidentifier The series UPFC branch neuro-identifier (SENI) in Figure 7 is a three layer feedforward neural network with thirteen inputs, a single hidden layer with fifteen sigmoidal neurons and two outputs. There are four different types of inputs, the first two types are the differences between the follow- ing signals: the measured real power and its reference value - Pen, and, the measured reactive power and its reference value Qerr. The other two types are the training signals - AEdgrbs and AEqprbs (Figures 8 and 9). In the pre- training phase, PRBS are applied to excite all possible dy- namics of the plant 1141. These PRBS are fed to the series inverter at B and SENI at C with the switches S1 and S2 at position 2 (Figure 3). Typical PRBS signals applied are shown in Figures 8 aind 9. A E d nrbs, A E o orbs 7 I 'Irl - ' 'Scrics 17 Njuroidcntificr a i Figure. 7 Neuroidentifier for series branch. ' I -0 32 I 1 1 4 ~ - i244 1348 1452 i s k s 166 Figure 8. PRBS signal Ed applied to series branch Time (sec) of UPFC. 246 IClSlP 2004 L 1 1 . 5 1 2 . 4 6 1 3 4 2 14.38 15.34 76.3 Figure 9. PRBS signal E, applied to series branch Time (sec) of UPFC. The frequency contcnt of this signal is lHz, 3Hz and 5Hz. The high magnitude of perturbation is required to have better identification of the system dynamics. All the four different types of inputs are time delayed (TDL) by onc sample period and together with their eight previously de- layed values form the twelve inputs to the SENT. The out- puts of the SENI at E are the estimated difference in the real power - I; err and in the reactive power - Q err at the next time step. The outputs of SENI at E are compared to outputs of the plant at D and the error signals at F are used to updatc the weights of the SENI using the backpropaga- tion algorithm. This process is repeated until a satisfactory error goal is obtained with the SENI leaming over a num- ber of different possible operating points of the plant. SIMULATION RESULTS The power system model consists of two synchronous gen- erators GI and GZ of ratings PI = 1600MVA and P2 = 2200MVA respectively, two loads one of value P (real power) =3000 MW, Q (reactive power) = 1800 W A R and another of value P = 3000 MW, Q = 300 WAR connected at Busz, and five transmission lines connected between buses 2 and 3. The UPFC is installed on line 5 of between buses 2 and 3. The inputs for the shunt branch control are deviation signals of VI (Voltage at the point of contact of the shunt branch) and Vdc (DC voltage across the capacitor) with their steady state reference values. The series branch is controlled by the deviation signals of Pinj and Qinj (real and reactive power) with their reference values respec- tively. This section of the paper mainly deals with the iden- tification of these deviation signals Verr and Vdcerr on the shunt branch and Perr and Qerr on thc scrics branch of the UPFC. Identification of Series and Shunt Branches of UPFC: Identification of the Verr and Vdcerr of the SHNI arc car- ried out and the weights are updated continually. The two training signals fed to SHNI - AEpdgrbs and AEpclprbs are those shown in Figures 5 and 6. Figures 10 and 12 show the comparison outputs of SHNI ( few(t) and Vd,,, ( t ) ) and the plant (Verr and Vdcerr). Figures 1 1 and 13 are the enlarged version of small sections of Figures 10 and 12 respectively. I I 44 53 - 4 4 M - 44 i s 44 sf 44 98 45 09 Figure I O . Actual signal Verr and identified signal of the shunt branch. I f e r 1 'Q - * 4486 4491 4496 4472 4477 4482 Time (sec) Figure 11. A section of actual signal Verr and identified signal fe, shown in Figure I O . Figure 12. Actual signal Vdcerr and identified sig- nal ?dcerrof the shunt branch. 1 Y - I -5 2361 I -- L- - V, 44 72 4477 4482 4486 - - 44 91 44 96 Time (sec) Figure 13. A section of actual signal Vdcerr and identi- fied signal fdcerrshown in Figure 12. Similarly the identification of the deviation signals Perr and Qerr by the SENI is carried out at the same time as that of SHNI and the weights are updated continually. The two training signals fed to SENI - AEdqrbs and AEqpbs are those shown in Figures 8 and 9. Figures 14 and 16 show the comparison of the outputs of SENI ( and Qerr(') ) and the plant (Perr and Qerr). Figures 15 and 17 247 IClSlP 2004 are the enlarged version of small sections of Figures 13 and 16 respectively. t2615 _-__ - _ _ -- ..____ Figure 14. Actual signal Perr and identified signal err of the series branch. L __I ~- . . . - . L-..-i_.- _i 44.62 44.68 44.74 44.81 44.87 44.93 Time (sec) Figure 15. A section of actual signal P,, and identified signal kw shown in Figure 14. 7 W . J i I I 44.51 44.61 44.92 45.02 Figure 16. Actual signal &,and identified signal Q of the series branch. CONCLUSION: Identification of shunt and series branches of UPFC plays an important role for the successful implementation of con- trol. In this paper, muroidentifiers that can estimate the outputs of the shunt and series branches of UPFC one step ahead accurately are proposed. These neuroidentifiers aid in the design of neurocontrollers, which are versatile in controlling the UPFC at various operating points. Further, these neuroidentifiers learns the dynamics of the shunt and series branches very fast, which is very important for real time online applications. Future work aims at design of a neurocontroller which can be used in effective control of a UPFC REFERENCES: Mathur, R. M., Varma, R. K., Thyristor-Based FACTS controllers for Electrical Transmission Systems, By IEEE Press and Jlohn Wiley & Sons, Inc, ISBN 0-471- Chunlei, L., Hongbo, S., Yu, D.C., “A novel method of power flow analysis with unified power flow con- troller (UPFC)”, IEEE Power Engineering Society Winter Meeting, vol. 4,2000, pp. 2800 -2805. Hingorani, N. G., Gyugyi, L., Understanding FACTS Concepts and Technology of Flexible AC Transmission Systems, Power Electronics Sponsored By: IEEE Power Engineering Society, 1999, ISBN 0-7803-3455- 8. Venayagamoorthy, G . K., Harley, R. G., “Two sepa- rate continually online-trained neurocontrollers for ex- citation and turbine control of a turbogenerator”, IEEE Transactions on Industry Applications, vol. 38, no. 3, MayfJune 2002, lpp. 887 -893. Dash, P. K., Mlishra, S., Panda, G., “A radial basis function neural network controller for UPFC”, IEEE Transactions on Power Systems, vol. 15, no.. 4, 2000, Crow, M. L., Dong, L. Y.; Zhang, L, “A new control strategy for the unified power flow controller”, IEEE Power Engineering Society Winter Meeting, 2002, pp. Venayagamoorthy, G. K., Harley, R.