IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO.

4, OCTOBER 2008

2255

Direct Frequency Domain Computation of Transmission Line Transients Due to Switching Operations
Adam Semlyen, Life Fellow, IEEE, and Abner Ramirez, Senior Member, IEEE
AbstractRealistic transmission lines are most easily represented directly in the frequency domain. For the computation of single switching operations to be started off in the frequency domain, switches have traditionally been modelled through the use of compensatory voltage/current sources. The practical computation of the complete transient response involves the calculation of the response for each switching operation for the whole observation time; sequentially, all of the individual responses are added up. This paper presents a novel frequency-domain-based methodology for calculating in a straightforward manner the transient response of transmission lines/underground cables due to switching operations. The core of the proposed methodology consists in using a fast Fourier transform-generated switching matrix where the on/off times are included. The computational stability of the transient response has been assured via the numerical Laplace transform which is free of the well-known numerical errors incurred by the direct use of a discrete Fourier transform. Index TermsAdmittance matrix, Fourier transform, frequency domain, switching operations, transmission lines.

I. INTRODUCTION RANSMISSION lines have to often withstand the effect of transients due to breaker operations. The resulting overvoltages may put on risk insulators, equipment, and personnel. Therefore, safety and economical reasons make it desirable to accurately calculate such overvoltages. The computation of transient waveforms during switching operations can be performed either in the time domain (TD) or in the frequency domain (FD). The former is a natural domain to observe time-dependent variables. Consequently, several TD programs have been developed and are widely used for calculating the time domain response of transmission lines under switching operations, for instance, the Electromagnetic Transients Program (EMTP) [1]. However, FD representation handles naturally frequency-dependent devices in a precise manner. In the case of transmission lines/underground cables, there is no need for dealing with rational approximations, time-delay identication, frequency-dependent matrices, and so on.
Manuscript received June 5, 2007; revised August 7, 2007. Current version published September 24, 2008. Paper no. TPWRD-00329-2007. A. Semlyen is with the Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON M5S 3G4, Canada (e-mail: [email protected]). A. Ramirez is with CINVESTAV-Guadalajara, Guadalajara 45010, Mexico (e-mail: [email protected]). Digital Object Identier 10.1109/TPWRD.2008.2002651

Several studies have reported on FD computations of transient waveforms during switching operations [2][5]. In these works, the basic idea is to replace the switch by a compensatory voltage/current source with its value calculated from the pre-closed/preopen state of the network. Then, by inserting those sources, the obtained switch action responses are added to the preceding ones. This can be done under the assumption of the system under study being linear. A switching-based piecewise approximation can be used for the representation of nonlinear loads [3], [4]. It is important to note that for each state of the switch, the system response must be calculated for the whole observation time. This could become a disadvantage of the existing methods when several switching operations are to be simulated. This paper presents a method for obtaining directly from FD the transient response due to switching operations of a transmission line (or system). It is based on the idea that the relation between a (discrete) TD variable and its FD representation via a fast Fourier transform (FFT) [6] can be formulated as a multiplication by a matrix . This then permits compact algebraic manipulations leading directly to a frequency domain admittance matrix (FDAM). The main underlying assumption is that the switching instants are xed to discrete times. This is a restriction in those important cases when the switching-off instant has to be at the natural zero crossing of the current. Since reverting from FD to TD by direct application of a discrete Fourier transform (DFT), of which the FFT is a particular case, results in numerical errors, such as Gibbs phenomenon and leakage (aliasing), in this study the use of the numerical Laplace transform (NLT) is proposed to avert those errors. Basically, the NLT is a modication of the DFT through function windows (Gibbs phenomenon) and a stability factor (aliasing) [7]. The development of the NLT and its application to transient analysis has been well documented in the last 40 years as can be seen in [8][19]. Further justication for using the NLT is the fact that this paper focuses on the transient waveforms resulting from switching rather than on periodic waveforms (i.e., steady state). In the former, Gibbs and leakage play an important role. A technique using the harmonic domain (based on a vector/matrix form of the Fourier coefcients) for calculating steady state, including switching devices, is described in [20]. The following sections are discussed. Section II describes the basic theory related to the representation of switches in FD. Section III examines some of the numerical issues related to the proposed methodology. Section IV presents an application example and Section V outlines further computational issues.

0885-8977/$25.00 2008 IEEE

2256

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 4, OCTOBER 2008

II. REPRESENTATION OF A SWITCH IN THE FREQUENCY DOMAIN A. Assumptions In this study, the switch is viewed as an ideal device with the following assumptions: and , are 1) the switching-on and switching-off times xed (i.e., they are independent of the values of the currents and voltages in the circuit); and are at discrete values of time, from an equidis2) tant set, corresponding to the discrete Fourier transforms (DFTs, in the guise of FFTs and IFFTs) used in the derivations. B. Linear Algebra of a Switch We adopt a multidimensional geometric view of the voltage and current vectors related to a switch. These contain elein FD. ments and are denoted generically by in TD and The time step in is over the total observation time , with samples from to . and are related by DFTs. In the Matlab The arrays implementation, the corresponding (very efcient) FFT and inverse fast Fourier transform (IFFT) are the functions fft and ifft (in this paper, we use FFT and DFT interchangeably). Their actions are linear operations that can be expressed by the and as follows: matrices (1a) (1b) The functions fft and ifft can act on arrays of vectors, in particular, on the identity matrix . Thus, (1a) and (1b) yield (2a) (2b) Of the two matrices, and , we will need and calculate only . That is, even though the computations in (2) are perdoes not have to be calformed very efciently, dened as culated via (2a). It is simply proportional (with before) to the complex conjugate of

The rst partitioning in (5) separates groups of rows, the second separates columns. Since the product of the matrices and gives the identity matrix , we obtain from (5), taking (3) into account

(6) so that and (7)

Since the inner product of two complex column vectors involves the transpose of the rst and the conjugate of the second, the equivalent relations (7) tell us that the columns forming the and and, thus, the respective subspaces and matrices any vectors that they would contain, are mutually orthogonal. (or ) is in the In other words, a vector in the range of (or ). These remarks will serve to characnullspace of terize the FD vectors of voltage and current and in terms of the subspaces to which they are conned. Now assume that a switch is closed over time segments toof the total observation time , and is accordingly taling open during . For the convenience of presentaportions at the betion and derivations, we shall lump the part at the end. Then, for the current ginning of and the , expressed as in (1b) in terms of , we have (8) or (9) Thus, is in the range of [see (7)] (10) so that (11) where is an arbitrary vector of appropriate length. From (9), we see that is in the nullspace of (12) across the switch, the situation and the For the voltage results are the duals of those for ; therefore, just the indices on and off have to be interchanged. We will need the dual of (9) (13) Up to this point, the subscript has been used for emphasis on FD variables. In the following, however, the variables will

(3) Both matrices and are symmetrical (but not Hermitian) (4) In the following, we will consider partitioned according to the time segments when the switch is on or off:

(5)

SEMLYEN AND RAMIREZ: DIRECT FREQUENCY-DOMAIN COMPUTATION OF TRANSMISSION-LINE TRANSIENTS

2257

generally be in FD, so will be dropped and only will be kept in the case of TD variables. Thus, (11) and (13) will now become

(14) (15)
Fig. 1. Single switch in an RL branch.

III. NUMERICAL ISSUES A. From DFT to NLT For convenience, we rewrite the transformation (1b) as (16) It is well known that the direct use of (16) leads to numerical errors due to Gibbs and leakage (aliasing) phenomena [7], [17]. Alleviating those errors can be done through the use of window functions (e.g., Hamming, Blackman, Von Hann) and a damping constant, respectively [7], [15], [19]. The modied version of (16), known as the numerical Laplace transform (NLT), can be expressed (see Appendix A) as (17) where is the discrete array of time, corresponds to the function window, and is a damping constant. In this study, the value (18) is used [19]. Therefore, an FD variable can be converted to TD in a precise manner (free of the errors mentioned before) by the use of the NLT expressed by (17). Although a detailed analysis of the transition from DFT to NLT is not the objective of this work, some basic equations have been given in Appendix A. We mention that although we use the / operators along this paper, in the computational implementation, they should be operators viewed as ifft/fft operations instead. The (which correspond to the exact transformation pair DFT/IDFT) are used along the calculations of the FDAM and the NLT is applied only when converting the resultant function (voltage or current) from the frequency domain to time domain by using (17). Errors of the order of 10 have been reported in the literature [19] when using the NLT as in (17). B. Illustrative Example As a preliminary example, consider a single branch with a switch as shown in Fig. 1. The voltage across the switch is (19)

and then (14) gives (22a) or (22b) Thus, we have obtained the FDAM of the system given by (23)

It is important to mention that usually we do not evaluate (22b) by using (23) as a matrix-vector multiplication sequence but by applying ifft operations (from right to left) wherever possible. The latter speeds up the computation of (22b) in cases (i.e., the number of columns in (5) correwhere is larger than the one corresponding to ). sponding to Similarly, the inverse of the matrix in (23) is calculated by trioperator from angular factorization by using the backslash Matlab. Fig. 2 shows the comparison between the use of (16), corresponding to the direct application of the FFT, and (17) using the NLT. Additionally, the TD numerical solution of the ODE given by the circuit in Fig. 1 is also shown in Fig. 2, although the difference between this solution and the NLT is not noticeable. The parameters (see Fig. 1) are , and rad/s; samples have been used for the three methods, namely, FFT, NLT, and the numerical solution of the ODE. For illustration, the switch is closed at the voltage zero-crossing at s and opened at s (a bit farther than a zero-crossing). From Fig. 2(b) [a close-up of Fig. 2(a)], the waveform showing numerical oscillations corresponds to the results given by (16) (DFT) and the smooth ones to the results by (17) (NLT) and the numerical solution of the ODE. Notice that the numerical oscillations have been diminished considerably through the NLT. Notice also that, as expected, no (Gibbs) oscillations result from the TD numerical solution. C. Properties of the FDAM

and the constraint (15) becomes (20) Substituting from (14) into (20) yields (21)

We note the following properties of the FDAM, of (23). It is neither symmetrical nor Hermitian since is, in general, complex. Its rank is only , where is the number of columns in [the central part in its expression (23) is an matrix, ]. Therefore, is not invertible. It means rather than

2258

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 4, OCTOBER 2008

Fig. 3. Single-line diagram for a three-phase open-ended transmission line.

The calculation of the FDAM for this system is described as follows. From the circuit, we have the voltage across the switch given by (24) where is the source impedance and represents the voltage at the sending end of the line (node 1 in Fig. 3). On the other hand, the transmission line is represented by (25) and , with , where and the characteristic admittance, the propagation matrix, and the length of the line, respectively. In this paper, the function of is calculated as where a matrix is a diagonal matrix with the eigenvalues of . We also note that it is convenient and reliable to use the general Matlab matrix function funm (see help funm) or to obtain the matrix hyperbolic functions via simple combinations of the matrix exponential function expm. Since the receiving end of the line is open ended, (25) yields
Fig. 2. Voltage and current waveforms for the RL branch.

(26a) If the line is terminated with a load sponding expression for becomes (see Fig. 3), the corre(26b) The substitution of (26a) or (26b) into (24) leads to (27) where (28) A similar process as in (20)(23) is applied to (27) to calculate the FDAM for the system shown in Fig. 3. The numerical results will be given. It is assumed here that the three phases of the receiving end of the line are permanently short-circuited by a resistance of 0.01 . For the purpose of illustration, the sequence of operations for the three-phase breaker is close-open-close-open. The , and is assumed to be at 180 , 120 , rst closing of phases and 60 , respectively. The rest of the operations are assumed to be with intervals of 360 . Since the fault is permanent, the nal state of the switch is open. For the numerical implementation, assume that the function samples are used to calwindow by Von Hann and culate the transient waveforms, with a total observation time

that while any set of harmonic voltages can be applied to , we cannot force an arbitrary set of harmonic currents through the circuit of Fig. 1 (i.e., currents that are not consistent with the switching). has a chessboard (sparsity) pattern in the sense that the even and odd harmonics in and are decoupled. (This is based on the assumption that the switching of has half-wave symmetry). In closing this section, we note that of (23) is a very compact result, involving only simple matrix operations. These are entirely in FD and there are no iterative procedures involved. All of the switching operations can be included in the switching matrix for the whole observation time. The computer implementation is very simple and straightforward. In the case of analyzing transients superimposed on a steadystate solution, one can make use of the superposition principle (i.e., the total response is given by the steady-state response plus the transient waveform). IV. APPLICATION EXAMPLE A. Energization of a Three-Phase Line Consider an open-ended three-phase transmission line as shown in Fig. 3 and with the conguration in Appendix B.

SEMLYEN AND RAMIREZ: DIRECT FREQUENCY-DOMAIN COMPUTATION OF TRANSMISSION-LINE TRANSIENTS

2259

equal to 70 ms (thus, from (18) ). Figs. 4(a)(c) show the voltages at nodes 1 and 2, and the current entering the line at node 1. For validation purposes, we have simulated the same system in PSCAD [21] by using the universal line model [22] and with the same number of samples. The results from the proposed methodology (continuous line) and the ones given by PSCAD (dashed line) are very close. Further simulations (not included here) have shown that the results from PSCAD get closer to the ones from the proposed methodology as the number of samples is increased in the former. Similarly, the procedure outlined before has also been successfully applied for a transmission system involving underground cables. V. FDAM AS A TIME-VARYING MATRIX In linear time-invariant (LTI) circuits, we are used to the idea that the admittance is independent of the phase of the input . However, since and dene the operational conditions relative to the phase of the applied voltages, the FDAM should . be viewed as a time-dependent operator In the FD, we can express the time-shifted (in the positive -direction by a time ) voltage/current vectors as (29) where the complex frequency is given by Substituting and from (29) into (22b) yields [19].

(30) where (31) is time dependent or, more preThis means that indeed cisely, it is function of the time reference . Any of its modied elements are given by (32) Expression (32) shows that the modication does not affect and, in general, the modication of any of the diagonal of its elements can be viewed as element-by-element multiplication by the (Hermitian) Toeplitz matrix with elements . When has to be modied due to a shift of the reference, related perhaps to some control action, the advantage of (32), versus recalculating as in (22a), is obvious. branch example from SecFor illustration, consider the tion III-B. While the source is kept the same along the observation time as in Fig. 2(a), the switch is now closed/opened with s. The source voltage and the cura time shift of rent for boththe original closing/opening times and the new onesare presented in Fig. 5. The time for calculating (32) in this example is about 8 times smaller than the one by using (22b). In the authors experience, that ratio usually falls between 8 and 10.

Fig. 4. (a) Voltage at node 1. (b) Voltage at node 2. (c) Current entering the line from node 1.

VI. DISCUSSION The NLT technique presented in this paper uses regular sampling, and the utilization of the FFT speeds up the solution dramatically. Some algorithms using nonuniform sampling have been proposed before [16], [18], [23]. Although it has been claimed that such procedures could save computing time, using them can become very cumbersome when multiple switching occurs. In fact, the most general sampling is the regular one;

2260

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 4, OCTOBER 2008

Fig. 6. Transmission-line conguration for the example in Section IV-A.

Alternatively, with
Fig. 5. Voltage and current waveforms for the RL branch including shifting.

, (33a) can be expressed as (33b)

it does not require knowing in advance the behavior of the transient (i.e., the fast dynamics at the beginning followed by the slow dynamics). Besides, the existing techniques handle switches as an initial condition problem [2][5]. Thus, the superposition principle has to be used, thereby making computation times excessive when multiple switching operations are simulated. We note that the proposed technique can be adapted for using odd sampling in the frequency domain; however, this issue is relegated for future work. VII. CONCLUSION In this paper, we have described a direct frequency domain methodology for calculating switching operations in transmission systems. The most attractive characteristic of the proposed methodology is that it calculates, at once, any number of switching operations in the frequency domain without the need to go back and forth to the time domain every time switching is performed in the system. The fact that we have used an FFT-based formulation permits a fast calculation of the transient waveforms. Additionally, a minor modication of the transfer function permits obtaining multiple simulations quickly. The described frequency-domain switch algebra is potentially applicable for power-electronic devices. Its presentation is planned in a forthcoming paper. The main difference from the periodic case which we intend to publish next is that in this paper, the transient is formulated in an articially periodic setting with the period resulting from a truncated observation time (truncated from innite to nite). Consequently, in this paper, the NLT is an essential part of the computational procedure. APPENDIX A be a transient waveform and its FD image (in the Let -domain). This pair of functions (rather than arrays of discrete values) is related by the inverse Laplace transform as follows:

From (33b), one can observe that the Laplace transform is equivalent to the Fourier transform of the attenuated signal . Since is a real and causal function, the inverse Laplace transform can be written as (34) Our objective is to obtain a numerical formula that permits us to evaluate (34) for any FD function , possibly transcendental or given in tabular form. This is attained through the discretization of (34) by the rectangular integration as described as follows. The discretization of (34) requires sampling the FD and TD and , respectively, as transient waveforms with intervals follows: for and (35a)

(35b) in (35b), we have arNote that for the discretization of samples corresponding to positive freranged the rst quencies (including zero frequency) followed by their ipped conjugate (excluding the ones at zero and maximum frequency). For the numerical integration of (34), we dene as the maximum frequency [15], [18], [19]. Then, with (36) rectangular integration of (34), using the denitions from (35) and (36), gives

(37) which becomes (17). Note from (37) that the discretized funchas been multiplied by the factor which corresponds tion to a window function [7].

(33a)

SEMLYEN AND RAMIREZ: DIRECT FREQUENCY-DOMAIN COMPUTATION OF TRANSMISSION-LINE TRANSIENTS

2261

APPENDIX B The conguration of the 200-km three-phase horizontal line used as an example in Section IV-A is shown in Fig. 6. The ground resistivity used is 100 m.

REFERENCES
[1] H. W. Dommel, Electromagnetic transients program reference manual (EMTP Theory Book), Prepared for Bonneville Power Administration. Portland, OR, 1986. [2] L. M. Wedepohl and S. E. T. Mohamed, Multiconductor transmission lines. Theory of natural modes and Fourier integral applied to transient analysis, Proc. Inst. Elect. Eng., vol. 116, no. 9, pp. 15531563, Sep. 1969. [3] N. Nagaoka and A. Ametani, A development of a generalized frequency domain transient programFTP, IEEE Trans. Power Del., vol. 3, no. 4, pp. 19962004, Oct. 1988. [4] P. Moreno, R. de la Rosa, and J. L. Naredo, Frequency domain computation of transmission line closing transients, IEEE Trans. Power Del., vol. 6, no. 1, pp. 275281, Jan. 1991. [5] P. Moreno, P. Gomez, J. L. Naredo, and J. L. Guardado, Frequency domain transient analysis of electrical networks including non-linear conditions, Int. J. Elect. Power Energy Syst., no. 27, pp. 139146, Feb. 2005. [6] W. L. Briggs and V. E. Henson, The DFTAn Owners Manual for the Discrete Fourier Transform. Philadelphia, PA: SIAM, 1995. [7] J. G. Proakis and D. G. Manolakis, Digital signal processing, in Principles, Algorithms and Applications, 3rd ed. Upper Saddle River, NJ: Prentice-Hall, 1996. [8] S. J. Day, N. Mullineux, and J. R. Reed, Developments in obtaining transient response using Fourier transforms. Part I: Gibbs phenomena and Fourier integrals, Int. J. Elect. Eng. Educ., vol. 3, pp. 501506, 1965. [9] S. J. Day, N. Mullineux, and J. R. Reed, Developments in obtaining transient response using Fourier transforms. Part II: Use of the modied Fourier transform, Int. J. Elect. Eng. Educ., vol. 4, pp. 3140, 1966. [10] S. J. Day, M. J. Battisson, N. Mullineux, and J. R. Reed, Developments in obtaining transient response using Fourier transforms. Part III: Global response, Int. J. Elect. Eng. Educ., vol. 6, pp. 259265, 1968. [11] N. Mullineux and J. R. Reed, Developments in obtaining transient response using Fourier transforms: Part IV-Survey of the theory, Int. J. Elect. Eng. Educ., vol. 10, pp. 2592, 1973. [12] L. M. Wedepohl and S. E. T. Mohamed, Transient analysis of multiconductor transmission lines with special reference to nonlinear problems, Proc. Inst. Elect. Eng., vol. 117, no. 5, pp. 979988, Sep. 1970. [13] L. M. Wedepohl and D. J. Wilcox, Transient analysis of underground power transmission systems,System model and wave-propagation characteristics, Proc. Inst. Elect. Eng., vol. 120, no. 2, pp. 253260, Feb. 1973.

[14] A. Ametani, The application of the fast Fourier transform to electrical transient phenomena, Int. J. Elect. Eng. Educ., vol. 10, pp. 277287, 1973. [15] D. J. Wilcox, Numerical Laplace transformation and inversion, Int. J. Elect. Eng. Educ., vol. 15, pp. 247265, 1978. [16] A. Ametani and K. Imanishi, Development of exponential Fourier transform and its application to electrical transients, Proc. Inst. Elect. Eng., vol. 126, no. 1, pp. 5156, Jan. 1979. [17] L. M. Wedepohl, Power system transients: Errors incurred in the numerical inversion of the Laplace transform, in Proc. 26th Midwest Symp. Circuits Systems, Aug. 1983, pp. 174178. [18] D. J. Wilcox and I. S. Gibson, Numerical Laplace transformation and inversion in the analysis of physical systems, Int. J. Numer. Methods Eng., vol. 20, pp. 15071519, 1984. [19] A. Ramirez, P. Gomez, P. Moreno, and A. Gutierrez, Frequency domain analysis of electromagnetic transients through the numerical Laplace transform, presented at the IEEE General Meeting, Denver, CO, Jun. 2004. [20] T. Noda, A. Semlyen, and R. Iravani, Entirely harmonic domain calculation of multiphase nonsinusoidal steady state, IEEE Trans. Power Del., vol. 19, no. 3, pp. 13681377, Jul. 2004. [21] Manitoba HVDC Research Centre Inc. Winnipeg, MB, Canada, PSCAD v.4.1.0. [22] A. Morched, B. Gustavsen, and M. Tartibi, A universal model for accurate calculation of electromagnetic transients on overhead lines and underground cables, IEEE Trans. Power Del., vol. 14, no. 3, pp. 10321037, Jul. 1999. [23] B. Gustavsen, Validation of frequency-dependent transmission line models, IEEE Trans. Power Del., vol. 20, no. 2, pt. 1, pp. 925933, Apr. 2005. Adam Semlyen (LF97) was born in 1923 in Rumania. He received the Dipl. Ing. degree from the Polytechnic Institute of Timisoara, Timisoara, Romania, in 1950, and the Ph.D. degree from the Polytechnic Institute of Iasi, Iasi, Romania, in 1965. He began his career there with an electric power utility and held academic positions at the Polytechnic Institute of Timisoara. In 1969, he joined the University of Toronto, Toronto, ON, Canada, where he is a Professor in the Department of Electrical and Computer Engineering, emeritus since 1988. His research interests include steady-state and dynamic analysis as well as computation of electromagnetic transients in power systems.

Abner Ramirez received the B.Sc. degree from the University of Guanajuato, Mexico, in 1996, the M.Sc. degree from the University of Guadalajara, Guadalajara, Mexico, in 1998, and the Ph.D. degree from the Center for Research and Advanced Studies of Mexico (CINVESTAV), Guadalajara Campus, in 2001. He was a Postdoctoral Fellow in the Department of Electrical and Computer Engineering at the University of Toronto, Toronto, ON, Canada, from 2001 to 2005. Currently, he is a Professor at CINVESTAV-Guadalajara. He is a member of the Mexican Association of Professionals and Students A.C. (PLAPTSAC). His interests are electromagnetic transient analysis in power systems and numerical analysis of electromagnetic elds.