1 Kalman Filter Based Detection and Mitigation of Subsynchronous Resonance with SSSC Thirumalaivasan Rajaram, Member, IEEE, Janaki Muneappa Reddy, Member, IEEE, and Yunjian Xu, Member, IEEE Abstract—In a long transmission line the use of static series synchronous compensation (SSSC) with fixed series capacitor enables fast control of power flow. There is a potential risk of subsynchronous resonance (SSR) due to the series capacitor. In this paper, we propose to use the Kalman filter (KF) for state estimation of subsynchronous components present in series compensated line and the mitigation of SSR. This novel Kalman filter based SSR damping controller is referred to as Kalman filter-damping controller. The design of Kalman filter-damping controller is based on the magnitude of damping torque in the range of torsional mode frequencies. We apply the genetic algorithm to optimize the controller parameters. The proposed Kalman filter-damping controller is highly frequency selective and effectively suppresses subsynchronous components of line current in the presence of close frequency oscillations. Analytical and simulation results at various system operating conditions demonstrate the effectiveness and robustness of the proposed Kalman filter-damping controller in mitigating SSR. Index Terms—Voltage Source Converter (VSC), Subsynchronous Resonance (SSR), Damping torque, Eigenvalue, Static Synchronous Series Compensator (SSSC), Kalman Filter (KF), Genetic Algorithm (GA). I. I NTRODUCTION T HE use of fixed series capacitor in electric power transmission line significantly increases transient and steadystate stability limits in addition to reactive power control. It is worth noting, however, that subsynchronous resonance is caused by interaction between two subsystems of the power systems, i.e., the electrical subsystem consisting of series compensated transmission lines, and the mechanical subsystem consisting of multimass turbine-generators. If controlling measures are not incorporated, the interaction of these two subsystems can aggravate the oscillations of generator rotors at subsynchronous frequency and result in generator shaft failures [1]-[4]. There exists a substantial literature on the design and analysis of countermeasures to mitigate the subsynchronous resonance (SSR) and to improve system stability. The improved blocking filter is designed to suppress different SSR problems [5]. The subsynchronous damping controller (SSDC) This work was supported in part by the Singapore University of Technology and Design through SRG project SRES13068. (Corresponding author: Yunjian Xu.) R. Thirumalaivasan is with the School of Electrical Engineering, VIT University, Vellore, India and doing post doctoral research in Engineering Systems and Design Pillar, Singapore University of Technology and Design, Singapore. M. Janaki is with the School of Electrical Engineering, VIT University, Vellore, India. Y. Xu is with the Engineering Systems and Design Pillar, Singapore University of Technology and Design, Singapore. (e-mail: yunjian xu@sutd.edu.sg). with the family of subsynchronous dampers (SSD) is proposed to provide positive damping by injecting damping currents into the generator [6]. Several other techniques have been proposed to alleviate SSR using FACTS (Flexible AC Transmission Systems) devices, for instance, the subsynchronous damping controller with STATCOM which provides positive damping in the range of critical torsional mode frequencies [7] and the supplementary controller with SSSC which adjusts the gain and phase of modal speeds to improve the damping of torsional mode [8]. Previous research has proposed several estimation algorithms that extract the subsynchronous components so as to effectively mitigate SSR with hybrid compensation consisting of fixed capacitor and FACTS devices (such as STATCOM or SSSC). To estimate the subsynchronous voltage components, the authors of [9] propose the low pass filters (LPF) estimation algorithm which is shown to provide more freedom for bandwidth selection than the recursive least square (RLS) algorithm. The suitability of two ambient algorithms for online monitoring of SSR modes, i.e., the frequency domain decomposition (FDD) and the RLS algorithm, are tested in [10], where both algorithms are shown to have good performance on tracking the changes in damping levels of the torsional modes and therefore can provide early alarms. Reference [11] addresses the mitigation of SSR using SSSC with an estimated subsynchronous voltage component in which low-pass filters are used in estimation algorithm. In our previous work [12], a subsynchronous current suppressor with band-pass filter is proposed to extract the subsynchronous components of line current, based on which the SSSC injects a proportional subsynchronous voltage to suppress the subsynchronous frequency current flowing through the generator and to mitigate SSR. In [13], Thyristor controlled series capacitor (TCSC) with Kalman filter is used to eliminate subsynchronous oscillations caused by SSR in power systems. In this paper, we propose an estimation algorithm that extracts the subsynchronous components of line current using Kalman filter. To mitigate SSR, SSSC injects a subsynchronous voltage in proportion to the estimated subsynchronous components of line current. The Kalman filter is a highly frequency selective, recursive model based least square estimator, which is widely used for state estimation of harmonics in power systems. This novel approach using Kalman filter is referred to as Kalman filter-damping controller. In what follows we discuss the advantage of the proposed approach over those used in the literature. The selection of bandwidth of filters is crucial so as to enhance the damping of torsional modes [9], [12]. In addition, 2 the accurate estimation of subsynchronous components requires low bandwidth filters with high selectivity. The Kalman filter does not need the selection of bandwidth required by low-pass filters [11] or the knowledge of eigenvalue analysis to choose the pass band for band-pass filters [12]. The resonance frequencies of (turbine-generator) mechanical systems are well known, hence the design of Kalman filter extracting subsynchronous components of torsional frequencies is simple as it merely relies on the knowledge of the torsional frequency. Since the Kalman filter is highly frequency selective, the proposed estimator is expected to provide more accurate estimates on subsynchronous components than low-pass and band-pass filters with low bandwidth, and to further improve the damping of torsional modes. To our knowledge, this work is the first that uses voltage source converter (VSC) based FACTS controllers with Kalman filter to mitigate SSR in power transmission networks. This paper presents SSR characteristics of series hybrid compensated power system (i.e., a combination of fixed capacitor and SSSC) using linear analysis and transient simulation. SSSC is based on a 3-level 24 pulse voltage source converter [14] and Type-1 controller [15], [16]. The detailed analysis of SSR is carried out with IEEE First Benchmark Model [17]. Based on the D-Q model, we conduct the damping torque analysis and eigenvalue analysis, and design the Kalman filter damping controller. The results of our linear analysis are validated using transient simulation with both the D-Q and a more detailed nonlinear three phase models of SSSC. We investigate the frequency sensitivity and selectivity of the proposed Kalman filter-damping controller in the presence of very close frequency oscillations. Our analytical and simulation results show the effectiveness of the proposed Kalman filterdamping controller in improving the damping of torsional modes and mitigating SSR. The rest of the paper is organized as follows. In Section II we present the modelling of the system. In Section III we present a case study that motivates the application of Kalman filter to mitigate SSR. In Section IV, we design the Kalman filter-damping controller and conduct performance evaluation. Some brief concluding remarks are given in Section V. Vg θg I Eb 0 + X Rl Generator X sys Xc l Rt X t VSC SSSC bc + gc a) Electrical system ω Te IP HP LPA LPB EXC GEN b) Six mass mechanical system Fig. 1. Modified IEEE First Benchmark Model with SSSC network is described by the following equation [4] xL I˙d Vgd RL ω̄XL Id = + Vgq −ω̄XL RL Iq ωB I˙q i (1) VD VcD EbD + T + + , VQi VcQ EbQ where T = cosδ sinδ −sinδ cosδ . Here, RL and XL are the series resistance and inductive reactance seen from the generator terminals. ω̄ is the per unit speed of the generator rotor. Vg , I, Vc , V i and Eb are generator terminal voltage, line current, capacitor voltage, SSSC injected voltage, and infinite bus voltage, respectively. The modeling of six-mass mechanical system, generator (2.2) model are detailed in [4], [18]. B. Modeling of SSSC in D-Q frame The schematic representation of SSSC and its phasor diagram are shown in Fig. 2, where Rs and Xs are the resistance and reactance of the interfacing transformer of VSC. In this paper, the SSSC is based on 3-level, 24-pulse voltage source converter [19], [20]. D II. M ODELING OF THE SYSTEM VR i VD i V = K mVdc i φ The IEEE first bench mark (FBM) model [17] is adapted in our study: the overall system consists of a turbine, a generator (2.2 model), as well as SSSC in series compensated long transmission lines, as shown in Fig. 1. The modeling of different subsystems are presented in the next a few subsections. Rs Vi φ Xs γ i γ VP VQ φ Q I A. State Model of Electrical and Mechanical System Fig. 2. Schematic and Phasor representation of SSSC. The analysis of subsynchronous resonance require the detailed modeling of the mechanical system in addition to the electrical system. In Fig. 1, the external network consists of a generator that is connected to infinite bus through series compensated long transmission line with SSSC. The external The D and Q components of SSSC injected voltage are described as, VDi = Km Vdc sin(φ + γ), (2) VQi (3) = Km Vdc cos(φ + γ). 3 Here, Km is the modulation index [19] and it is given as kρcosβ for 24-pulse three-level converter, which is a function of dead angle β. ρ is the transformation ratio of SSSC interfacing transformer. In Fig. 2, the phase angle of converter voltage leads the phase angle of line current φ by γ. In the right subplot of Fig. 2, VP is the real voltage of V i and in phase with line current I, whereas the quadrature term VR is the reactive voltage. The relation between the real and reactive voltages (VP and VR ) is VR = VDi cos φ − VQi sin φ, (4) VP = VDi sin φ + VQi cos φ. (5) III. A CASE STUDY In this case study, we consider the following initial operating conditions and assumptions. 1) The generator output power (Pg ) is 0.9 p.u. 2) The turbine (mechanical) input power is taken constant. 3) The study is carried out for two cases with 0.76 p.u total series compensation. In Case-1, 0.76 p.u series compensation is provided by fixed capacitor (Xc ) alone. In Case-2, series hybrid compensation is used with 0.25 p.u of compensation by SSSC (XSSSC = VR /I) and 0.51 p.u of compensation by fixed capacitor Xc . 4) In transient simulation, a step change of 10% decrease in mechanical input torque is applied at 0.5 sec and it is restored at 1 sec. A 3-phase to ground fault at generator terminal (with fault impedance given in [17]) is applied at 1 sec and cleared after 3 cycles. The positive values of VP and VR show active power absorption from the line and the inductive mode of operation of SSSC, respectively. The detailed modeling of SSSC can be found in [12]. A. Eigenvalue analysis C. SSSC Voltage control In a 3-level converter topology, Type-1 controller is used to control the magnitude of the VSC output voltage by varying the dead angle β with fundamental switching frequency [21] and the phase angle by varying γ. The voltage of DC side capacitor (Vdc ) is maintained constant by controlling real voltage VP and hence the set value of real voltage VPord is obtained from the DC voltage controller. The set value of reactive voltage VRord can be kept constant or obtained from a power scheduling controller. The eigenvalues of the system matrix for both cases are given in Table I. We note that the torsional mode-1 becomes unstable in Case-1, and the inclusion of SSSC (Case-2) decreases undamping of mode-1. As the frequency of network mode (sub) closely matches with torsional mode-2, it is found to be unstable in Case-2. It is clear that the application of SSSC increases and shifts the network resonant frequency. TABLE I E IGENVALUES OF THE COMBINED SYSTEM WITH AND WITHOUT SSSC. Torsional Mode Vdc 0 1 2 3 4 5 Network mode(sub) Network mode(super) 1 1 + s Tmd − + + kp ord γ and β + − VPsub VP Σ γ calculator VR ki s ord + β Σ − sub VR −4.9272 ± j 99.378 −3.7700 ± j 128.410 −5.7326 ± j 658.090 −4.8842 ± j 590.200 Type-1 controller for SSSC. The pictorial representation of Type-1 controller for SSSC is given in Fig. 3. It is worth noting that the voltages VPsub and VRsub are obtained from the Kalman filter-damping controller and are used to modulate the real and reactive voltage references of SSSC. γ and β are given by VR(ord) , (6) γ = tan−1 VP (ord) q β = cos−1 2 VP2(ord) + VR(ord) kρVdc . (7) 80 Rotor angle in deg Fig. 3. Case 2 : W ith SSSC (Xc = 0.51 p.u and XSSSC = 0.25 p.u) −1.7105 ± j 8.2100 0.0977 ± j 99.080 0.1578 ± j 127.000 −0.6670 ± j 160.460 −0.3790 ± j 202.850 −1.8504 ± j 298.170 LPA-LPB section torque in p.u Vdcref Σ Case 1 : W ithout SSSC (Xc = 0.76 p.u) −2.6346 ± j 9.3403 2.8086 ± j 97.473 −0.0807 ± j 126.960 −0.6582 ± j 160.490 −0.3740 ± j 202.850 −1.8504 ± j 298.170 75 70 65 0 2 4 6 Time in sec 8 10 0.8 0.75 0.7 0.65 0.6 0.55 0 2 4 6 Time in sec 8 10 Fig. 4. Response of rotor angle and LPA-LPB section torque for 10% decrease in input mechanical torque with three phase model of SSSC. 4 B. Transient simulation In MATLAB-SIMULINK [22], the transient simulation with combined system including SSSC (Case-2) in both D-Q and 3-phase model is carried out for mechanical input torque disturbance. The responses are shown in Fig. 4, and the increase in oscillations of LPA-LPB section torque indicates that the system is unstable. C. Damping Torque Analysis with linearized Model of SSSC The stability in torsional mode can be evaluated using damping torque analysis [7]. Variation of damping torque is shown in Fig. 5 for Case-1 and case-2. It is worth noting that without SSSC (Case-1), the maximum negative damping occurs at about 98 rad/sec and matches with mode-1 frequency. Hence, severe torsional interactions are expected. In Case2 (with SSSC), the peak negative damping is significantly reduced and shifted. Hence undamping of mode-1 reduces; however subsynchronous electrical frequency is close to 127 rad/sec and the negative damping is substantial. As a result, the torsional mode-2 is expected to be unstable. This is consistent with our previous eigenvalue analysis. 10 0 Case - 2 with SSSC TDe (p.u) -10 -20 Case - 1 without SSSC -30 -40 60 Fig. 5. 70 80 90 100 110 120 ωm (rad/sec) 130 140 150 160 Variation of damping torque with and without SSSC. It is straightforward to see from the damping torque analysis that the negative damping at critical torsional mode frequencies should be reduced to improve the torsional mode stability. The oscillations of turbine-generator at natural frequencies may cause subsynchronous current flow through the transmission network to the generator. Suppressing the subsynchronous current in transmission lines could help to improve the the damping in critical torsional modes. In the next section, we study the detection of subsynchronous components and the mitigation of SSR using Kalman filters. IV. K ALMAN F ILTER DAMPING C ONTROLLER AND A NALYSIS OF SSR The Kalman filter is commonly used for estimation of state variables in power systems. The design of Kalman filter is simple as it merely depends on torsional mode frequencies, and the resonance frequencies of the mechanical system are well known. In this paper, Kalman filter is used to extract subsynchronous frequency components in line current. SSSC is controlled to inject a proportional subsynchronous voltage to suppress the subsynchronous frequency current flowing through the line and the generator, and to improve the damping in the range of critical torsional mode frequencies. We propose a new approach to reduce subsynchronous currents flowing through the transmission line and the generator with a properly designed Kalman Filter-Damping Controller. The state estimation of subsynchronous current and the structure of Kalman filter-damping controller will be introduced in the next a few subsections. A. Subsynchronous Current Estimation with Kalman Filter Suppose that the generator rotor oscillates (at about a constant speed of ωo ) sinusoidally. If the amplitude of rotor oscillation is small, the induced voltage in the generator stator consists of three sinusoidal components in α-β sequence [4] as follows eα (t) = sup efα (t) − esub α (t) − eα (t), eβ (t) = sup efβ (t) + esub β (t) + eβ (t), (8) where ef (t), esub (t), and esup (t) are the fundamental, subsynchronous, and supersynchronous components of induced voltage in stator, respectively. In general the supersynchronous component results in positive damping, and subsynchronus results in negative damping. As such, when the voltage given in Eq. (8) is applied to the external network, the subsynchronous currents flow through the network depends on network impedance and is given by sub [Iqsub + jIdsub ] = Z −1 [j(ωo − ωm )]e−j(ωo t+δ) [esub β + jeα ], (9) where 1 Z[j(ωo − ωm )] = RL + j (ωo − ωm )L − (ωo − ωm )C (10) is the impedance (at the subsynchronous frequency) of the external network in Fig. 1; here, L includes the transient inductance of the generator. According to transformation matrix, the subsynchronous current in D-Q frame is given by sub sub [IQ + jID ] = ejδ [Iqsub + jIdsub ]. (11) The magnitude of a subsynchronous frequency component is given by q sub )2 + (I sub )2 . I sub = (IQ (12) D The induced voltage in the generator is given in α-β sequence in Eq. (8). The Kalman filter is designed to extract the subsynchronous components from the D-Q components of line current. In (9)-(12), we derive the expressions for the D-Q components of line current. We establish the relation between the state of the Kalman filter and the D-Q components of a subsynchronous frequency current (that is to be filtered) as sub T x1 xsub xsub xsub 2 3 4 (13) sub T sub sub sub ImD⊥ ImQ† ImQ⊥ = ImD† , where the symbols † and ⊥ indicate in-phase and quadrature components of the estimated subsynchronous current, m is the torsional mode, and xsub and xsub are the states corresponding 1 2 5 to in-phase and quadrature components of subsynchronous fresub quency current ImD from ID , respectively. Similarly xsub and 3 sub x4 are the states corresponding to in-phase and quadrature sub components of subsynchronous frequency current ImQ from IQ , respectively. Hence we consider four state variables for sub each subsynchronous frequency current Im . The state space model of Kalman estimator for a subsynchronous frequency current is defined as sub sub x1 x1 w1 1 0 0 0 0 1 0 0 xsub xsub 2 + w2 , 2sub = 0 0 1 0 xsub x3 w3 3 sub sub w4 k 0 0 0 1 k x4 x4 k k+1 (14) xsub 1 xsub 2sub + v1 , x3 v2 k k xsub 4 k y1sub y2sub = k H1 Z Z H2 Z sin(ωm tk ) cos(ωm tk ) = 0 0 R 0 i=j i6=j At every iteration, we compute the following Kalman filter matrix equations. • Prior estimation of state and error covariance: − sub [xsub k+1 ] = Axk , − [Pk+1 ] = APk AT + Q. • Compute Kalman gain: Kk = Pk− H T (HPk− H T + R)−1 . Update estimate: sub + sub − − [xsub − H[xsub . k ] = [xk ] + Kk yk k ] Update error covariance: Pk+ = (I − Kk H)Pk− . where Pk , Kk , and I are the error covariance matrix, Kalman gain, and the identity matrix, respectively. • B. Structure of Kalman Filter-Damping controller , . Here, ωm and tk are torsional mode frequency and the time at sub sub iteration, respectively. ImD and ImQ are the subsynchronous components from ID and IQ , respectively. As shown in Fig. 1, the IEEE FBM with six mass mechanical system has five natural torsional mode frequencies [4]. The modal inertia of torsional mode-5 is very high, and hence the Kalman filter is intended to extract four subsynchronous frequency components for torsional modes m=1, 2, 3 and 4. Combining the state equations of the four subsynchronous frequency currents, we define the state space model of Kalman estimator as sub xsub k+1 = Axk + wk , yksub = Hk xsub k + vk , (15) where H1k = H2k = E{vi vjT } = The structure of Kalman filter-damping controller is shown in Fig. 6. The extracted subsynchronous components from the Kalman filter are passed through appropriate gains k1 to k8 to sub sub , which are summed up to obtain signals and VmQ obtain VmD sub sub VD and VQ . The output of KF-damping controller are X sub VDsub = VmD , VQsub = m X The output of KF-damping controller VDsub and VQsub (in D-Q frame of reference) are then transformed to in-phase and quadrature components VPsub and VRsub respectively, and are used to modulate in-phase and quadrature voltage orders VPord and VRord of the SSSC as shown in Fig. 3 in section-2C. sub sub (16) k1 sub I1Q (17) where wk and vk are the process and observation noise vectors, respectively; their covariance matrices are given by Q i=j T E{wi wj } = 0 i6=j (19) m I1D where dim xsub = 4m × 1 and dim yksub = 2m × 1. k sub sub T sub xsub = [xsub 1 , x2 , ... x15 , x16 ]k are the states of the k subsynchronous current components corresponding to four torsional modes, and yksub = [y1sub , y2sub , ... y7sub , y8sub ]Tk is the measurement vector which is the output (i.e., the extracted subsynchronous current components) of Kalman filter. A and Hk are the state transition and measurement matrices respectively. We are now ready to describe the relation between the output of Kalman filter and the D-Q components of filtered subsynchronous frequency currents: sub sub T y1 y2 ... y7sub y8sub sub sub sub sub sub sub sub T = isub , (18) 1D i1Q i2D i2Q i3D i3Q i4D i4Q sub VmQ . sub V1D Σ sub k2 VD sub V1Q Σ VQ sub I2D sub I2Q ID Kalman Filter IQ (output) y sub Σ k4 sub I3D sub I3Q I4D Σ k6 sub k7 sub I4Q Σ k5 sub Fig. 6. Σ k3 V4D sub k8 V4Q Block diagram of Kalman filter-damping controller. The performance of Kalman filter depends on the observation and process noise covariance. The effect of increasing process noise covariance Q increases the error in the measured value. It is therefore important to tune the observation noise covariance R and gains k1 to k8 to improve the damping of 6 98.6 Mode-0 10 Imag part Imag part 15 Rlocus-1 Rlocus-2 5 0 -1.5 98.3 -1 -0.5 Real part 0 98.2 -1 -0.8 -0.6 Real part 220 Mode-3 and Mode-4 Mode-2 126.986 Imag part Imag part Rlocus-1 Rlocus-2 98.4 126.988 126.984 Mode-1 98.5 Rlocus-1 Rlocus-2 126.982 200 Rlocus-1 (mode-3) Rlocus-1 (mode-4) Rlocus-2 (mode-3) Rlocus-2 (mode-4) 180 160 126.98 -0.098 -0.096 -0.094 -0.092 -0.09 Real part 140 -0.7 -0.6 -0.5 -0.4 Real part -0.3 Fig. 7. Root locus-1 for variation in gains k1 to k8 (from 1 to 5) and root locus-2 for variation in observation noise covariance (from 1 to 5). C. Optimization of Kalman filter-damping controller Parameters The Kalman filter (KF)-damping controller is designed to enhance the damping of critical torsional modes. Genetic algorithm (GA) [23] is adopted to optimize the gains k1 to k8 of the KF-damping controller. To improve the damping of critical torsional modes, we seek to minimize the deviation between the desired damping torque TDe(des) and actual damping torque TDe in the range of torsional mode frequencies. The robustness of KF-damping controller is achieved by incorporating variant compensation levels in the optimization. The optimization problem is formulated as, XX 2 minimize SSE = TDe(des) − TDe (ω) , (20) Xc ω subjected to the constraint that the real parts of all eigenvalues are negative (to ensure the stability of the system). In (20), SSE is the summation of squared errors over the range of series compensation (Xc = 0.05 to 0.75 p.u and Xsssc = 0.25 p.u, and 50 ≤ ω ≤ 300 rad/sec) up to 100%. The desired damping torque is taken as 8 p.u, to ensure that the real parts of all eigenvalues are negative. 5 with SSSC and KF-Damping controller 0 TDe (p.u) -5 -10 with SSSC -15 -20 -25 50 100 150 200 250 300 ωm (rad/sec) Fig. 8. Damping torque with SSSC and Kalman filter-damping controller. D. Analysis of SSR with KF-damping controller The SSR with KF-damping controller is evaluated using eigenvalue analysis, damping torque analysis, and transient simulation. A graphical portrait of reactance with SSSC and KF-damping controller is presented in this section to illustrate the resonance condition. We further find out that the solution of the GA optimization (k1 to k8 ) remains unchanged at various operating conditions, which demonstrates the robustness of the proposed KF-damping controller. 1) Damping Torque Analysis: Fig. 8 shows how the damping torque changes with frequency ωm . We observe that the peak negative damping is significantly reduced with GA optimized KF-damping controller. With the KF-damping controller, negative damping in the range of torsional frequencies (60 − 300 rad/sec) is negligible. As a result, the system is expected to be stable with the intrinsic mechanical damping. Real part of eigenvalues of torsional modes torsional modes. Fig. 7 shows how the eigenvalues of combined system (SSSC with KF-damping controller) change with varying observation noise covariance R and gains k1 to k8 . Root locus-1 is computed using varying gains k1 to k8 (from 1 to 5) and fixed observation noise covariance. Root locus-2 is computed using varying observation noise covariance (from 1 to 5) and fixed gains k1 to k8 . It is noticed that the trajectories of eigenvalues in root locus-1 and root locus-2 are moving closely in same direction. We observe that the damping of torsional modes depends on the observation noise covariance; also, varying the gains k1 to k8 has more significant effect on the damping of eigenvalues of torsional modes than varying the observation noise covariance. Therefore we tune k1 to k8 to improve the damping of torsional modes while keeping the observation and process noise covariances fixed. We follow the systematic approach proposed in [12] to tune the gains k1 to k8 . In the following subsection, we apply the genetic algorithm to optimize the parameters in the Kalman filterdamping controller based on damping torque. 0.05 Mode-5 Mode-2 0 Mode-4 -0.05 Mode-3 -0.1 -0.15 -0.2 Mode-1 -0.25 -0.3 Mode-0 -0.35 -0.4 0.3 0.4 0.5 0.6 0.7 0.8 Total compensation (X c+Xsssc) 0.9 1 Fig. 9. Real part of eigenvalues of torsional modes for variation in compensation level with Kalman filter-damping controller. In Fig. 9, we plot the real parts of eigenvalues for all torsional modes over the entire compensation range from 0.3p.u to 1p.u. (the mechanical damping is neglected). We observe that all torsional modes are stable over the entire compensation range, i.e., the system is robust against damping subsynchronous oscillations under the practical range of series compensation. 7 3 Xc+Xse 2 Xc X se 1 XL 100 150 200 250 Electrical resonance frequency (ωm) in rad/sec Fig. 10. 300 80 Graphical portrait of reactance. 2) Graphical portrait of resonance condition: A graphical portrait of reactance with SSSC and KF-damping controller (Case-2) to illustrate the resonance condition is depicted in Fig. 10. It shows the variation of capacitive reactance Xc , inductive reactance XL , emulated reactance Xse of SSSC on single phase basis, and total effective capacitive reactance (with the KF-damping controller) Xc + Xse are plotted as a function of the frequency ωer [12]. The total effective capacitive reactance Xc + Xse never equals XL in the frequency range of 50-300 rad/sec, and as a result, the proposed KFdamping controller ensures that the series compensated power system is free from SSR. 3) Eigenvalue Analysis: The eigenvalues of the combined system with SSSC and KF-damping controller are shown in Table II. TABLE II E IGENVALUES OF THE COMBINED SYSTEM WITH SSSC AND K ALMAN FILTER - DAMPING CONTROLLER Torsional mode 0 1 2 3 4 5 W ith SSSC and KF − dampingcontroller (Xc = 0.51 p.u and XSSSC = 0.25 p.u) −1.4476 ± j 8.3088 −0.4579 ± j 98.369 −0.0892 ± j 126.980 −0.6864 ± j 160.520 −0.4053 ± j 202.870 −1.8504 ± j 298.170 Network mode subsynchronous −33.039 ± j 44.836 Network mode supersynchronous −66.983 ± j 469.660 We note from Fig. 10 that with Kalman filter damping controller the effective capacitive reactance (Xc + Xse ) is increased. Since the resonance frequency (ωer ) is proportional to the capacitive reactance, the resonance frequency (ωer ) is expected to increase. From Fig. 10, we observe that the subsynchronous resonance does not occur in the frequency range of 50-300 rad/sec with Kalman filter damping controller. Thus in Table II, the frequency of the subsynchronous network mode (ωer ) is reduced as compared to Table I (Case-2). Comparing the eigenvalue results in Table II (SSSC with KF-damping controller) to column-2 of Table I (SSSC without Rotor angle in deg 0 50 75 70 65 0 5 10 0.8 0.75 0.7 0.65 0.6 0.55 15 0 5 Time in sec 10 Time in sec 15 Fig. 11. Response of rotor angle and LPA-LPB section torque for 10% decrease in mechanical input torque with the SSSC and the Kalman filterdamping controller that is activated at t = 5 sec. 4) Transient Simulation: Fig. 11 shows the transient simulation of the combined model with SSSC and KF-damping controller under a full load of Pg = 0.9 p.u. We observe that as a result of mechanical disturbance the section torque increases with time. When the KF-damping controller is activated at t = 5 sec, the oscillation of shaft section torque decays with time. The FFT analysis of the LPA-LPB section torque is shown in Fig. 12, from which we observe that the mode-1 component is increasing and becomes predominant in 3-5 sec time span. When the KF-damping controller is activated at t = 5 sec, the mode-1 component decays quickly with time, demonstrating the effectiveness of the KF-damping controller to suppress the subsynchronous frequency components in transmission lines. Absolute magnitude Reactance in p.u 4 KF-damping controller), we conclude that the KF-damping controller leads to 1) significant enhancement in the damping of eigenvalues of torsional modes 1 and 2, 2) marginal enhancement in the damping of eigenvalues of torsional modes 3 and 4, 3) marginal reduction in the damping of eigenvalue of torsional mode-0, 4) significant increase in the damping of subsynchronous network mode, 5) and no effect on torsional mode-5. LPA-LPB section torque in p.u 5 0.1 0.1 0.05 5 - 6 sec 0 0 100 200 300 6 - 7 sec 0.05 0.05 0.05 0 0.1 0.1 4 - 5 sec 3 - 4 sec 100 200 300 0 100 200 300 100 200 300 Frequency (rad/sec) Frequency (rad/sec) Frequency (rad/sec) Frequency (rad/sec) Fig. 12. FFT analysis of LPA-LPB section torque with Kalman filter-damping controller (activated at t = 5 sec). Transient simulation for three phase fault with Pg = 0.9 p.u and Pg = 0 p.u are carried out and shown in Fig. 13. It is observed that with the KF-damping controller the oscillations of section torque decay with time. Fig. 14 shows the magnitudes of the line current along with subsynchronous currents and the D-Q components of total subsynchronous current under the three phase fault (with Pg = 0 p.u). The simulation result demonstrates the effectiveness of the Kalman filter-damping controller in extracting and suppressing the subsynchronous frequency components even when the line current (at the fundamental frequency) is zero as Pg = 0 p.u. 8 for Pg = 0.9 p.u 0 -5 -10 5 10 Time in sec 15 I sub 1 sub I1 5 2 1 0 0.1 0 I I sub 1 I sub 2 I sub 3 I sub 4 I sub D Time in sec 10 -2 0 5 0.2 -4 0 5 10 Time in sec 15 2 1 0 0.1 0 Fig. 15. 0.2 ×10 2 1 0 0.1 ×10-3 9.5 Time in sec 10 0 5 0.2 -3 2 1 0 0.1 10 5 Time in sec 10 10 0 ×10 Time in sec 10 -3 9.5 0 0 10 5 Time in sec 10 Response of the Kalman filter for different Q values. subplots in Fig. 15 show the measurement error in magnitude of subsynchronous components. Response-1 is for the actual values of Q used in simulation: Q1 = 3.5e−8 ; Q2 = 1e−8 ; Q3 = 1e−9 ; Q4 = 1e−9 ; 5 0 5 Time in sec 10 15 Q5 = 1e−9 ; Q6 = 1e−9 ; Q7 = 1e−9 ; Q8 = 1e−9 . Response-2 is for incremented values of Q: 0 5 Time in sec 10 15 Q1 = 3.5e−6 ; Q2 = 1e−6 ; Q3 = 1e−7 ; Q4 = 1e−7 ; 10 15 Q5 = 1e−7 ; Q6 = 1e−7 ; Q7 = 1e−7 ; Q8 = 1e−7 . 0 0 5 Time in sec 10 15 0 5 Time in sec 10 15 0 5 Time in sec 10 15 0 5 Time in sec 10 15 0 0.3 I sub Q 10 5 10 0 0.1 0.4 0 -0.4 0 ×10-3 9.5 0.1 0.05 9.5 0 0 0 10 ×10-3 Time in sec 0 0.1 0.1 Time in sec 10 sub 3 0.2 2 0.2 0.2 2 1 0 9.8 Time in sec 10 4 Fig. 13. Response of rotor angle and LPA-LPB section torque for 3-phase fault at generator terminal with SSSC and Kalman filter-damping controller for Pg = 0.9 p.u and Pg = 0 p.u. 0.2 0.2 sub 2 0 5 10 0 9.5 0.4 0 ×10-3 9.5 0 3 2 1 0 9.6 0 Time in sec 10 I 0.1 15 ×10-3 0 10 I 5 5 10 Time in sec 3 LPA-LPB section torque Rotor angle in deg 0 for Pg = 0 p.u 10 0 -2 15 2 1 0 5 I sub 4 5 10 Time in sec 9.8 5 0.2 I sub 0 0 2 50 0 0 0 Response-2 0.5 ×10-3 9.6 I sub 70 5 2 sub 90 Response-1 0.5 4 I4 LPA-LPB section torque Rotor angle in deg 110 0 -0.3 Fig. 14. Magnitudes of line current along with extracted four subsynchronous currents and D-Q components of total subsynchronous current for 3-phase fault at generator terminal when Pg = 0 p.u with Kalman filter-damping controller. E. Frequency Selectivity of Kalman Filter The state estimation of Kalman filter algorithm depends on the values of the process noise covariance (Q) and the observation noise covariance (R). Hence the frequency selectivity of Kalman filter can be illustrated through its response to different values of Q and R. In Section IV-B we discussed the effect of varying the observation noise covariance. Next, we present the effect of varying the process noise covariance for two sets of Q values. Fig. 15 shows the magnitude of subsynchronous components from the Kalman filter for two sets of Q values. The We observe from Fig. 15 that the effect of increasing process noise covariance (Q) increases the error in the measured value. We also observe that the overshoot and the response time are higher in Response-2. These results indicate that the frequency selectivity of Kalman filter primarily depends on the process noise covariance (Q) and the observation noise covariance (R). Since the time scale of the variance of Q and R is in general larger than the time scale of SSR, Q and R can be considered to be fixed in our analysis; their values are given in Appendix A. F. Effectiveness and sensitivity of KF-damping controller in close frequencies scenario We carry out detailed analysis of SSR for the proposed Kalman filter - damping controller on the IEEE FBM model, which is the standard benchmark model to test the effectiveness of any countermeasures in validating the damping of SSR. For a power plant consisting of several generators with near and spread torsional mode frequencies, multiple torsional mode frequency components are expected to present in the line current. In such situations, damping controllers should be designed to mitigate all critical torsional mode oscillations [24]. In multi-machine systems it is important to analyze the effectiveness and sensitivity of damping controller to close frequency oscillations. We have investigated the sensitivity of 9 KF- damping controller and analyzed the damping effectiveness in the presence of close frequency oscillations. In what follows, we present the investigation of damping effectiveness and sensitivity for the proposed Kalman filter - damping controller. 1) Effectiveness of KF-damping controller: We seek to investigate the damping effectiveness of the designed Kalman filter in the presence of close frequency oscillations. We consider a signal in the following form Ie (t) = Ae e−at sin(ωe t + φe ), (21) Absolute magnitude which is assumed to imitate an oscillation in line current, and is termed as simulating signal [24]. In this analysis, the simulating signals of 15.2, 19.7, 25 and 31.78 Hz are chosen to test the effect of near frequency oscillations. The effectiveness of Kalman filter - damping controller is tested for different amplitude of simulating signals, which are added at the time of fault clearance to the line current. The transient simulation of the combined system is carried out with three phase fault that is applied at t = 1 sec and is cleared after 3 cycles. FFT analysis of simulating signals for the time span of 1 sec are shown in Fig. 16. Fig. 17 demonstrate the effectiveness of Kalman filter damping controller through FFT analysis of Kalman filter output for different amplitudes of simulating signals. The simulating signals of different frequencies are expected to distort the actual subsynchronous component. We observe from Fig. 17 that the magnitude of subsynchronous components (output of the designed Kalman filter) is increased due to simulating signals and the response time of damping is significantly increased for torsional mode-2. However, as time progresses the magnitude of four subsynchronous components are reduced with Kalman filter - damping controller, indicating the stability of the system. These results demonstrate the effectiveness of the proposed KF-damping controller to suppress the subsynchronous frequency components in the presence of close frequency oscillations. 0.04 0.04 1 - 2 sec 0.03 0.02 0.01 0.01 0 0.02 150 200 0 100 simulating signals (of ampliude 0.1) simulating signals (of ampliude 0.2) Fig. 16. 3 - 4 sec 0 100 and vary the magnitudes of subsynchronous components. It is straightforward to see that the simulating signal frequencies are not present in the Kalman filter’s output, demonstrating that KF-damping controller is not sensitive to the close frequencies and has high frequency selectivity. 0.04 2 - 3 sec 0.03 0.02 Fig. 17. FFT analysis of Kalman filter output for different amplitudes of simulating signals. 150 200 100 150 200 Frequency (rad/sec) FFT of simulating signals. 2) Sensitivity of KF-damping controller to very close frequencies: The authors of [25] propose to apply damping controllers for multi-machine plants. Their field tests show that torsional modes frequencies of generators are very close. Based on this observation in [25], we investigate the sensitivity of the designed Kalman filter on 0.1 Hz close frequencies. We carry out the sensitivity analysis with the simulating signals at frequencies of 15.6, 20.1, 25.4 and 32.18 Hz. FFT analysis of Kalman filter output for the time span of 1 sec is shown in Fig. 18, from which we observe that the simulating signals distort subsynchronous components in line, Fig. 18. FFT analysis of Kalman filter output with (0.1 Hz) close frequency simulating signals. Remark 4.1: We propose a novel Kalman filter based subsynchronous damping controller to extract the subsynchronous frequency components from the line current. The design of KF-damping controller is simple as for any compensation level, Kalman filter extracts subsynchronous frequency components corresponding to torsional mode frequencies. The incorporation of the KF-damping controller improves the damping 10 of all torsional modes and eliminates SSR in the entire range of series compensation. The analytical and simulation results exhibit robust performance of the system under different operating conditions, and therefore demonstrate the effectiveness of the KF-damping controller in mitigating SSR. V. C ONCLUSION In this paper, we analyze the SSR characteristics of a hybrid compensated transmission line with series capacitor and SSSC. We propose a simple technique for the extraction of subsynchronous frequency components using Kalman filter. The design of the Kalman filter-damping controller is based on the magnitude of damping torque in the range of torsional mode frequencies. We apply the genetic algorithm to optimize the controller parameters. The results of various analysis demonstrate the following. 1) The inclusion of SSSC reduces the peak negative damping. 2) The incorporation of a suitably designed KF-damping controller significantly improves the damping in the entire range of compensation levels, under all critical torsional modes and different operating conditions. 3) Under various types of disturbances, the proposed KFdamping controller effectively extracts and suppresses the subsynchronous components of line current even when the fundamental frequency line current at the operating point is zero. 4) The proposed KF-damping controller completely eliminates the electrical resonance conditions as well as the SSR under practical series compensation levels. 5) The proposed KF-damping controller is highly frequency selective and effectively suppresses subsynchronous components of line current in presence of close frequency oscillations. A PPENDIX A PARAMETERS OF THE KF- DAMPING CONTROLLER R1 = 1.2; R2 = 5; R3 = 0.3; R4 = 0.3; R5 = 0.3; R6 = 0.3; R7 = 0.2; R8 = 0.2; Q1 = 3.5e−8 ; Q2 = 1e−8 ; Q3 = 1e−9 ; Q4 = 1e−9 ; Q5 = 1e−9 ; Q6 = 1e−9 ; Q7 = 1e−9 ; Q8 = 1e−9 . [6] L. Wang, X. Xie, Q. Jiang, and H. R. Pota, “Mitigation of Multimodal Subsynchronous Resonance Via Controlled Injection of Supersynchronous and Subsynchronous Currents”, IEEE Trans. on Power Systems., vol. 29, no. 3, pp. 1335-1344, May 2014. [7] K. R. Padiyar and Nagesh Prabhu, “Design and Performance Evalution of Subsynchronous Damping Controller with STATCOM”, IEEE Transactions on Power Delivery, vol. 21, no. 3, pp. 1398-1405, July 2006. [8] Dipendra Rai, Sherif O. Faried, G. Ramakrishna and Abdel-Aty (Aty)Edris, “An SSSC-Based Hybrid Series Compensation Scheme Capable of Damping Subsynchronous Resonance”, IEEE Transactions on Power Delivery., vol. 27, no. 2, pp. 531-540, April 2012. [9] Massimo Bongiorno, Jan Svensson and Lennart Angquist, “Online Estimation of Subsynchronous Voltage Components in Power Systems”, IEEE Transactions on Power Delivery, vol. 23, no. 1, pp. 410-418, Jan 2008. [10] H. Khalilinia and V. Venkatasubramanian, “Subsynchronous Resonance Monitoring Using Ambient High Speed Sensor Data”, IEEE Trans. on Power Systems, to be published. [11] M. Bongiorno, J. Svensson, and L. Angquist, “Single-Phase VSC based SSSC for Subsynchronous Resonance Damping”, IEEE Trans. on Power Delivery, vol.23, no.3, pp. 1544-1552, July 2008. [12] R. Thirumalaivasan, M. Janaki and Nagesh Prabhu, “Damping of SSR Using Subsynchronous Current Suppressor With SSSC”, IEEE Transactions on Power Systems, vol. 28, no. 1, pp. 64-74, Feb. 2013. [13] E. Gustafson, A. Aberg and K. J. Astrom, “ Subsynchronous resonance. A controller for active damping”, in Proc. of the 4th IEEE Conference on Control Applications, Sep. 28-29, 1995. [14] N. G. Hingorani and L. Gyugyi, Understanding FACTS, New York: IEEE Press, 2000. [15] Schauder and Mehta, “Vector analysis and control of advanced static VAR compensators”, IEE Proc.-c, vol. 140, no. 4, pp. 299-306, July 1993. [16] K. R. Padiyar and A. M. Kulkarni, “Design of reactive current and voltage controller of static condenser”, Int. J. Electr. Power Energy Syst., vol. 19, no. 6, pp. 397-410, 1997. [17] “First bench mark model for computer simulation of subsynchronous resonance”, IEEE Transactions on Power App. Syst., vol. 96, no. 5, pp. 1565-1572, sep/oct 1977. [18] K. R. Padiyar, Power System Dynamics - Stability and Control- Second edition, Hyderabad: B. S. Publications, 2002. [19] K. R. Padiyar and Nagesh Prabhu, “Analysis of subsynchronous resonance with three level twelve-pulse VSC based SSSC”, in Proc. IEEE TENCON-2003, Oct. 14-17, 2003. [20] J. B. Ekanayake and N. Jenkins, “Mathematical models of a three level advanced static var compensator”, IEE Proc.- Generation Transm. distrib, vol. 144, no. 2, March 1997. [21] K. R. Padiyar, FACTS controllers in Power Transmission and Distribution, New Delhi, India. New age International (P) Ltd, Publishers, 2007. [22] Using MATLAB-SIMULINK, The MathWorks, Inc., Natick, MA, 1999. [23] Goldberg, “Genetic Algorithm In search, Optimization and Machine Learning”, Addison Wesley Reading, 1989. [24] X. Xie, L. Wang, X. Guo, Q. Jiang, Q. Liu, and Y. Zhao, “Development and field experiments of a generator terminal subsynchronous damper”, IEEE Transactions on Power Electronics, vol. 29, no. 4, pp. 1693-1701, April 2014. [25] X. Xie, L. Wang, and Y. Han, “Combined application of SEDC and GTSDC for SSR mitigation and its field tests”, IEEE Trans. on Power Systems, vol. 31, no. 1, pp. 769-776, January 2016. R EFERENCES [1] M. C. Hall and D. A. Hodges, “Experience with 500 kV subsynchronous resonance and resulting turbine generator shaft damage at Mohave generating station”, in Analysis and Control of Subsynchronous Resonance, 1976, IEEE Publ. 76 CH 1066-O-PWR. [2] C. E. J. Bowler, D. N. Ewart, and C. Concordia, “Self excited torsional frequency oscillations with series capacitors”, IEEE Transactions on Power App. Syst., vol. PAS-92, pp. 1688-1695, 1973. [3] L. A. Kilgore, D. G. Ramey, and M. C. Hall, “Simplified transmission and generation system analysis procedures for subsynchronous resonance problems”, IEEE Transactions Power App. Syst., vol. PAS-96, pp. 1840-1846, Nov./Dec. 1977. [4] K. R. Padiyar, Analysis of Subsynchronous Resonance in power systems, Boston: Kluwer Academic Publishers,1999. [5] X. Xie, P. Liu, K. Bai, and Y. Han, “Applying Improved Blocking Filters to the SSR Problem of the Tuoketuo Power System”, IEEE Trans. on Power Systems, vol. 28, no. 1, pp. 227-235, Feb. 2013. R. Thirumalaivasan (M’12) received his Ph.D. degree in the Department of Electrical Engineering, JNTU Hyderabad and M.Tech degree from College of Engineering, Anna University, Guindy, Chennai. He is an associate professor in the School of Electrical Engineering at VIT University, Vellore, India and pursuing his post doctoral research in SUTD, Electrical Systems Design, Singapore. His research interests include FACTS, HVDC, and Real-time digital simulation of power electronics and power systems. 11 M. Janaki (M’12) received her Ph.D. degree in the Department of Electrical Engineering, JNTU Hyderabad and M.E degree from College of Engineering, Anna University, Guindy, Chennai. She is an Associate Professor in the School of Electrical Engineering at VIT University, Vellore, India. Her research interests include FACTS, HVDC, and power systems. Yunjian Xu (S’06-M’10) received the B.S. and M.S. degrees in electrical engineering from Tsinghua University, Beijing, China, in 2006 and 2008, respectively, and the Ph.D. degree from the Massachusetts Institute of Technology (MIT), Cambridge, MA, USA, in 2012. Dr. Xu was a CMI (Center for the Mathematics of Information) postdoctoral fellow at the California Institute of Technology, Pasadena, CA, USA, in 2012-2013. He joined the Singapore University of Technology and Design, Singapore, as an assistant professor in 2013. His research interests focus on energy systems and markets, with emphasis on power system control and optimization, wholesale electricity market design, and the aggregation of distributed energy resources in power distribution systems. Dr. Xu was a recipient of the MIT-Shell Energy Fellowship.