Saad Dilshad

Proc. of the 4rth International Conference on Power Generation Systems and Renewable Energy Technologies (PGSRET) 10-12 September 2018, Islamabad, Pakistan Improving the Stability of Power System using Type-II NeuroFuzzy Based Damping Control for HVDC Muhammad Abdul Basit Department of Electrical Engineering COMSATS University Islamabad, Pakistan- Rabiah Badar Department of Electrical Engineering COMSATS University Islamabad, Pakistan- Saad Dilshad Department of Electrical Engineering COMSATS University Islamabad, Pakistan- Abstract— High Voltage DC (HVDC) is now becoming a promising technology to deliver large amount of power to farthest areas of power network. Moreover, its application includes stability of surrounding grids when installed in conventional AC transmission grid. In this research, a Type-II Adaptive NeuroFuzzy Wavelet (ANFW) based damping control for HVDC has been introduced to improve the stability of overall system. This type of control enables better management of a complex system with uncertainties. This structure provides a dynamic membership function with more degree of freedom and makes it possible to grasp antecedent part uncertainties in a better way. Performance and efficiency of Type-II ANFWC is tested using chronological faults on Single Machine Infinite Bus (SMIB) system. Performance improvement of proposed controller is then investigated by comparison with conventional controls i.e. Artificial NeuroFuzzy Takagi Sugeno Kang Control (ANFTSKC) and Lead Lag Control (LLC). Commutation Converter (LCC) and Voltage Source Converter (VSC), depending upon the underlying technology. HVDC links can be installed in parallel with High Voltage AC (HVAC) transmission lines to enhance power system transient stability by damping LFOs [5]. Due to its economical, geographical and environmental advantages, HVDC is considered as more suitable choice as compared to HVAC transmission for bulk power transmission over long distances. A fast change of power flow through LCC based HVDC systems can be used to improve transient stability of entire system [6]. Keywords— type-II NeuroFuzzy control, wavelet, HVDC, low frequency oscillations, power system stability HVDC system along with different supplementary controllers reduce the effect of disturbance up to tolerable limits thus enhancing the overall stability of the system [8]. Therefore, PI controllers are designed to compute the power flow settings of HVDC system. Different AC variables are used as PI parameters [9]. In [10], a novel control based on Phasor Measurement Unit (PMU) is presented for HVDC. Imbalance in phase voltage angle of two areas is taken as input of PMU controller. In addition to that some nonlinear techniques like Model Predictive Control (MPC) have also been used in designing of supplementary damping control for HVDC [11]. However, these conventional and nonlinear controls have performance and complexity issues. On the other hand, adaptive NeuroFuzzy controllers have been found to better cope with stability issues of power system due to their learning and adaptive qualities [12, 13]. I. INTRODUCTION Electric power system is a large nonlinear complex system where technicians and engineers are continuously facing issues related to stability. Stability of a power system means that all machines of power system should operate in synchronism [1]. Low Frequency Oscillations (LFOs) originate in the system whenever any internal and external fault occurs. Insufficient damping torque of generating units is responsible for these oscillations. LFOs are further categorized as local mode and inter-area modes of oscillations. Local mode oscillations (ranging between 0.1-2 Hz) arise when a group of generators or a single generator swings against rest of the system. However, inter-area mode of oscillations (ranging between 0.1-0.8 Hz) result due to mismatch in swing between group of generators of multiple areas [2]. Different conventional methods have been used in literature to counter these LFOs such as Power System Stabilizer (PSS) and Automatic Voltage Regulator (AVR) [3]. Later, with advancement in power electronics based technology, controllers like first and second generation Flexible AC Transmission Systems (FACTS) are extensively being used to damp LFOs. They provide flexibility and stability by controlling reactive power with injection of voltage or current in power system [4]. Another rapidly growing technology, based on power electronics converters, is High Voltage DC (HVDC) transmission. HVDC systems are mainly categorized as Line -/18/$31.00 ©2018 IEEE There are three different configurations to integrate AC power with HVDC technology: (i) connect AC busses through HVDC cables (ii) HVDC grids embed into AC system (iii) DC segmentation. DC segmentation means that a large system is disintegrated into small segments and these small segments are connected via AC/DC links [7]. However, these Adaptive NeuroFuzzy (ANF) control strategies mostly have Type-I membership function in antecedent part following a linear consequent part in the conventional structure of Takagi-Sugeno-Kang (TSK). Thus, these types of ANF control strategies are vulnerable to uncertainties of a highly dynamic plant like power system. Different types of uncertainties are reported in literature which cannot be handled by Type-I fuzzy systems [14]. Therefore, Type-II fuzzy system was introduced for control of complex non-stationary plants [15]. Type-II fuzzy system can depict the system uncertainties with more precision in the rule base due to due to blurred and fuzzy boundaries of its membership functions. However, this type of network may get stuck in the local minima while searching for optimal solution due to inherent drawback of optimization techniques like gradient descent. This dilemma can be resolved by integrating wavelets in the conventional structure of TSK based NeuroFuzzy. Wavelet Neural Networks (WNN) learn the plant dynamics by adaption of wavelet coefficients which makes it highly suitable to use with adaptive control schemes [16]. Unlike our previous work [12], [13], this research presents the synergistic integration of Morlet wavelet based NN with Type-II fuzzy systems thus considering the improvement in both the antecedent and consequent part of NeuroFuzzy architecture. Main contribution of this research is to: iteration until LFOs damp completely. Average model of HVDC system is used in this work, where dependent voltage source powered by; Vd = U do cos α Where, U do = 3 2U sec V p.u . π ( N p / Ns ) (1) and U sec is secondary side voltage of transformer. By modeling all losses caused by resistance and thyristor’s switching, (1) is rewritten as: Vd = U do cos α − Rc I d − Req ( drop ) − 2V f (2) Where, V f is the fixed voltage drop across the thyristor, • design and implement Type-II ANFWC based LCCHVDC for transient and steady state stability improvement. Req is the voltage drop across the equivalent resistance and can be found as: • validate the effectiveness of Type-II ANFWC damping action by comparative evaluation with conventional control schemes using multiple fault scenarios at different loading conditions. Req (drop) = Req × I d Remainder of this paper is organized as follows: Closed loop system structure is described in section II. Detailed architecture and mathematical treatment of proposed control scheme is presented in section III. Section IV is composed of simulation results and discussion. In the end, section V summarizes the concluding remarks and future work. II. CLOSED LOOP SYSTEM STRUCTURE Fig. 1 shows the closed loop system structure, where SMIB system is installed with HVDC link. Error signal is fed to auxiliary damping controller. Technical specification of master control is specified by technical structure of project. Same amount of current signal I ord is sent to both rectifier and inverter controls to avoid the loss of margin. A reference ramp current Id ref with an adjustable time, ramp up or down is added to ignite or stop the converter working [17]. Reference current is ramped up to its final value when system get stabilized and ramped down to minimum value before stopping. Master control use telecommunication link to send current order signal to converter stations. That signal influences the firing angle α which is the only free control variable in CSC-HVDC. Controlled α put its impact on active power flow through HVDC link. α change in each Fig. 1. Closed-loop system structure (3) Extinction angle γ is calculated as: γ = 180 − α − μ (4) Rectifier and inverter control generates α R and α I , respectively, and used to calculate overlap angle μ .  2R I  μ = cos−1  cos α − c d  − a (5) U do   Demand of reactive power at both end of the converter control can be calculated as:  I q × U do × π × K  (6) Q = −  6   AC active power flowing into converter transformer is given as:  I q × U do × π × K  P=  6   Where, K = tap ratio × nom trans ratio (7) III. CONTROL ARCHITECHTURE AND DESIGN The Fuzzy Inference System (FIS) described completely in terms of Type-I fuzzy system is called Type-I FIS. ANFIS is the most popular model which combines the benefits of Adaptive Neural Network (ANN) and FIS into a single capsule. ANFIS implements a TSK-FIS. Adaptive NeuroFuzzy TSK Control (ANFTSKC) is also considered as Type-I NeuroFuzzy system because the membership function used in ANFTSKC is Gaussian membership function with linear function in its consequent part. Type-I fuzzy system cannot model some types of uncertainties due to rigid boundary of membership function. Type-II fuzzy can directly model these uncertainties because its membership functions are fuzzy in nature. Type-II fuzzy sets have blurred boundaries. The uncertainty in the primary membership grade of Type-II membership function consists of a bounded region, this bounded region is called Footprint Of Uncertainty (FOU). There are uncertainties associated with mean and Standard Deviation (STD) of the membership function. Gaussian membership function used in this work has fixed STD and uncertain mean. A. Proposed Type-II ANFW Control The generalized fuzzy rule used to describe the Type-II ANFWC structure is: h i =1 Here, x1 , x2 ,..., xh and χ1 , χ 2 ,..., χ g are the input and output variables, respectively. C denotes jth Type-II ji membership function for ith input such that j = 1, 2,3,, g . These rules are implemented in a layered structure, presented in Fig. 2. It consists of five consecutive layers; each layer has its own defined task. Inputs are taken into the controller by the nodes of layer 1, each node corresponds to the number of inputs to the system. Gaussian membership grade for upper and lower part is determined by layer 2, each layer contributes one linguistic term. The mathematical expression of Gaussian membership function with uncertain mean is:  1 ( xi − eij ) G ( eij , ςij , xi ) = exp  −  2 ςij2    , e ∈ ( e1 , e 2 ) (8) ij ij  ij  Where, e1 and e 2 are the two uncertain mean parameters and ς is the fixed STD. The upper and lower Gaussian membership functions based on uncertain mean are denoted as μ and μ , respectively or equivalently as μC ji and μC . Mathematically, these are given as: ji G(e1ij , ς ij , xi ), xi < e1ij  μij ( x) = 1, e1ij ≤ xi ≤ e2ij G(e 2 , ς , x ), xi > e2ij ij ij i  e1ij  G (e2ij , ς ij , xi ), xi ≤ μij ( x) =  G (e1 , ς , x ), x > e1ij ij ij i i  In layer 3, the membership degree of each rule is calculated using a T-norm product operator. The output of layer 3 for jth rule is; f j = μC ( x1 ) * μC ( x2 ) * ⋅⋅⋅ * μC ( xh ) j1 (9) 2 jh (11) f j = μC ( x1 ) * μC ( x2 ) * ⋅⋅⋅ * μC ( xh ) j2 jh Layer 4 is the first layer of consequent part. Each node has a Morlet based wavelet neural network which consists of three sub layers: input layer, hidden layer and output layer. as: The hidden layer contains Morlet wavelet functions given M j ( xi ) = cos(5 pij )e Where, pij = xi − τ ij d ij −0.5 pij (12) , dij and τ ij are the dilation and translation parameters of Morlet wavelet. The output of layer 4 is given by; h χ j = w j  M j ( xi ) (13) i =1 Type reduction and defuzzification is carried out in layer 5 and 6, respectively. Then, the following inference engine calculates the output of the Type-II NFW: g y = yType − II = q f j χ j j =1 N  g + (1 − q )  f j χ j j =1 N  fj j =1 (14) fj j =1 where, ‘y’ is the final output of the Type-II ANFWC and ‘q’ is the design factor. It determines the contribution of lower and upper membership functions in final output [16]. B. Adaptaion Mechanism of Proposed Type-II ANFWC The control system parameters are updated by the following update rule based on gradient decent: + e2ij 2 + e2ij j2 j1 IF x1 is C j1and x2 is C j 2and  xh is C jhTHEN χ j is w j  M j ( xi ) 2 Fig. 2. TYPE_II ANFWC architechture A ( n + 1) = A(t ) − γ (10) Here, ∂K ∂A A = ς ij , e1ij , e2ij , dij ,τ ij , w j , q  (15) is adaptive parameter vector and γ is the learning rate of the gradient decent update rule. The parameters of the Type-II ANFWC are updated by minimizing the following cost function: 1  2 K = ( zr − z ) + y 2 (16) 2 2 Where, zr and z are the reference and the actual output of the plant, respectively and y is the controller output. The partial derivatives in the (15) are calculated using following chain rule:  ∂y ∂f j ∂μ j ∂K ∂y ∂f j ∂μ j  = ν  +  ∂e2ij j  ∂f j ∂μ j ∂e2ij ∂f j ∂μ j ∂e2ij  ∂K ∂y ∂χ j ∂M j ∂pij = ν ∂dij ∂χ j ∂M j ∂pij ∂dij (17) (18) ∂K ∂y ∂χ j ∂M j ∂pij = ν ∂τ ij ∂χ j ∂M j ∂pij ∂τ ij (19) ∂K ∂y ∂χ j = ν ∂w j ∂χ j ∂w j (20) ∂K ∂y = ν ∂q ∂q (21) observed in SMIB system. Rotor speed deviation and power flow on AC line is shown in Figs. 3(a) and 3(b), respectively. It has been observed that Type-II AFNWC performs significant damping effect in both the transient and steady state regions. First swing transient stability improvement can be seen in second and third fault scenario. Proposed controller brings the system in steady state in least time for all faults. Injected current signal into master control is shown in Fig. 3(d), it is observed that Type-II ANFWC has smooth control effort as compared to other controls. B. Case-II (Light Loading) In this case, electrical power of generating unit is set at Pe = 0.4 p.u. Location, duration and time of fault occurrence of all the three faults are kept same as in first case. Rotor speed deviation, AC line power flow, rectifier current and control signal as injected current are shown in Fig. 4. Where,  ∂K ∂z  + y   ∂z ∂y  ν =  (a) Here, the term ∂z / ∂y is called plant sensitivity measure. It describes the change in plant output with respect to the change in control input. Plant sensitivity for the direct adaptive control can be set to a constant [20]. The complete update equations can be found using the values of (16) to (21) in (15). IV. RESULTS AND DISCUSSION A SMIB system is designed in MATLAB/ Simulink to validate the performance of Type-II ANFWC. 2100 kVA power generating unit generates 13.8 kV. The voltages are step up for transmission by a transformer. SMIB system is modified by installing an HVDC link in parallel with conventional double circuit transmission line. HVDC link operates at 500 kV and 2000 A with total capacity of 1000 MW. Converter transformers are connected at both converter ends. Length of each transmission line is 300 km. The system is installed with 250 MW load. Difference of actual and desired rotor speed deviation of synchronous generator is assigned as input to the damping controller. Type-II ANFWC, ANFTSKC and LLC are tested and compared using multiple faults at three different loading conditions on HVDC based power system. Details of LLC can be found in [18] and the choice is made through a selector switch. A. Case-I (Nominal Loading) The system discussed above is set at nominal loading by adjusting electrical power of generating unit Pe = 0.75 p.u. A 3 − ϕ fault for 5 cycles is applied on line L2. Fault is initiated at t = 0.1 sec. Second fault of 10% step increase in reference voltage is applied on generating unit at 3 sec. for duration of 1 sec. The third and the last fault is applied on DC line rectifier side at 6 sec. After every fault LFOs are (b) (c) (d) Fig. 3. Case-I: (a) Rotor speed deviation (b) AC power flow (c) Rectifier current (d) Control effort It has been observed, that post-fault oscillations amplitude is smaller in this case as compared to that of nominal loading scenario due to light loading condition. Fig. 4(b) shows that power flow on AC line has been reduced from 788 MW to 584 MW. Simulation results for system parameters shown in Fig. 4 reveal that although Type-II ANFWC maintains its superior performance in terms of both the transient stability and settling time, however, ANFTSKC has more competitive performance in this scenario as compared to nominal loading. C. Case-III (Heavy Loading) In order to validate the robustness of the proposed control scheme the system has now been exposed to more stressed scenario of heavy loading. Furthermore, the performance against change in fault location has also been investigated by changing the location of first fault from line L2 to line L1 and shifting the third fault from rectifier end of DC line to inverter side. Fault occurrence time and duration is same as in previous two cases. The power flow through AC lines is approximately 1100 MW. It has been found in literature that at heavy loading condition, SMIB system becomes unstable when subjected to large fault [21]. However, installation of HVDC link resolves this dilemma and system remains intact but with poorly damped oscillations. Fig. 5 shows that Type-II ANFWC performs significantly better as compared to other controls and quickly damps the oscillations in system parameters. (a) (a) (b) (c) (d) Fig. 4. Case-II: (a) Rotor speed deviation (b) AC power flow (c) rectifier current (d) Control effort (b) (c) (d) Fig. 5. Case-III: (a) Rotor speed deviation (b) AC power flow (c) Rectifier current (d) Control effort D. Quantitative Analysis Performance Indices (PIs) are calculated to compare the percentage improvement of the proposed controller. Formula used to calculate the results is given as: [2] [3] T PIs =  t m | Δω |n dt (22) 0 Where, ( m, n ) ∈ {( 0,1) , ( 0, 2 ) , (1,1) ,1, 2} for IAE, ISE, ITAE and ITSE, respectively. Results obtained using (22) are presented in Table I. It has been found that improvement margin for ITSE is greater as compared to other performance indices which shows that performance improvement in transient region is more significant in transient region as compared to steady state. Furthermore, ITAE, IAE and ISE for Type-II ANFWC is greater as compared to ANFTSKC showing its improved performance in steady state. TABLE I. [5] [6] [7] [8] PERFORMANCE IMPROVEMNT W.R.T. LLC [%] Controllers Case [4] Type-II ANFWC I II III ANFTSKC I II III [9] PIs IAE 41.77 30.22 43.57 25.21 19.90 26.19 ITAE ISE - - - - - - ITSE 40.07 76.37 61.38 25.58 19.83 29.89 V. CONCLUSION A Type-II ANFWC is proposed as supplementary damping control for LCC based HVDC link. In proposed controller, Type-II fuzzy membership functions are used in antecedent part and Morlet wavelet networks are used in consequent part of NeuroFuzzy controller which embeds the fuzziness property of type-II control and localization property of wavelets in NeuroFuzzy structure. SMIB system installed with HVDC link is subjected to series of fault under different loading conditions. It has been observed that installation of HVDC link keeps the system intact in the event of large fault but with poorly damped oscillations. However, inclusion of supplementary damping control helps to improve the damping performance. It is concluded that proposed Type-II ANFWC has significant performance improvement, especially, in more stressed scenarios. This paper investigates the controller performance for transient stability enhancement against local mode of oscillations. The controller application can further be extended to multimachine system to validate its performance for inter-area mode of oscillations. Moreover, the overall closed-loop strategy can be extended to indirect control scheme for further performance improvement but at the cost of complexity. [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] REFERENCES [1] B. K. Kumar, “Power system stability and control,” Academic press, Dept. Elect. Eng., IIT Madras, Chennai, India, 2004. Ayesha, et al., “An adaptive controller scheme implemented on multiple links VSC-HVDC for effective damping of low frequency power oscillations,” IEEE Int. Conf. Power Syst. Tech. (POWERCON). NSW Australia, pp. 28 Sep.-01 Oct. 2016. C. Liu, R. Yokoyama, K. Koyanagi, and K. Y. Lee, “PSS design for damping of inter-area power oscillations by coherency-based equivalent model,” Int. J. Electr. Power Energ. Syst., vol. 26, pp. 535544, September 2004. D. Murali and M. Rajaram, "Active and reactive power flow control using FACTS devices," Int. J. Comp. Appl., vol. 9, pp: 45-50, November 2010. K. P. Basu, “Stability enhancement of power system by controlling HVDC power flow through the same AC transmission line,” IEEE Symp. Ind. Electron. Appl. (ISIEA). Kuala Lampur, pp-663-668, October 2009. I. M. Sanz, B. Chaudhuri and G. Strbac, “Corrective control through HVDC links: a case study on GB equivalent system,” IEEE Power Energ. Soc. General Meeting (PES), Vancouver BC, July 2013. S. P. Azad, R. Iravani and J. E. Tate, “Damping low-frequency oscillations by tuning the operating point of a dc-segmented ac system,” IEEE Power Energ. Soc. General Meeting (PES), Vancouver BC, 2013. S. K. Kerahroudi, F. Li, M. M. Alamuti, R. Rabbani, G Taylor and M. Bradley, “Power system stability enhancement of the future GB transmission system using HVDC link” 49th Int. Uni. Power Eng. Conf. (UPEC). Romania, September 2014. J. Hazra, Y. Phulpin and D. Ernst, “HVDC control strategies to improve transient stability in interconnected power systems,” IEEE Power Tech. Bucharest, pp. 1-6, October 2009. Y. Huang and Z. Xu, “HVDC supplementary controller based on synchronized phasor measurement units,” IEEE PES Power Syst. Conf. Expo. NY USA, pp. 668- 672, October 2004. S. P. Azad, R. Iravani and J. E. Tate, “Damping inter-area oscillations based on a model predictive control (MPC) HVDC supplementary controller,” IEEE Trans. Power Syst., pp-, vol. 28, Aug 2013. J. Shair, M. A. Basit and R. Badar, “Damping low frequency oscillations in an HVDC system using Neuro-Fuzzy SMC,” IEEE Int. Conf. Power, Energ. Smart Grid (ICPESG). Mirpur Pakistan, pp- 1-6, April 2018. M. A. Basit, R. Badar and F. Jan, “Intelligent damping control for transient stability enhancement using HVDC,” 5th Int. Conf. Electr. Eng. (ICEE) Lahore, pp. 1-6, 2018. J. M. Mendel and R. B. John, “Type-2 fuzzy sets made simple,” IEEE Trans. Fuzzy Syst., vol. 10, pp. 117–127, 2002. R. H. Abiyev and O. Kaynak, “Type 2 fuzzy neural structure for identification and control of time-varying plants,” IEEE Trans. Indus. Electron., vol. 57, pp-, 2010. Y. Bodyanskiy and O. Vynokurova, “Hybrid adaptive wavelet-neurofuzzy system for chaotic time series identification,” Info. Sci., vol. 220, pp. 170–179, 2013. A. T. Rao, P. Ramana and P. K. Rao, “Modeling of cigre HVDC benchmark system in MATLAB/ SIMULINK,” Int. J. Edu. Appl. Res. (IJEAR), vol. 4, pp. 112-115, India, 2015. P. Sidhartha, “Robust coordinated design of multiple and multi-type damping controller using using differential evolution algorithm,” Int. J. Electr. Power Energ. Syst., vol. 33, pp-, 2011. T. Takagi and M. Sugeno, “Fuzzy identification of systems and its applications to modeling and control,” IEEE Trans. Syst., Man, Cyber., vol. SMC-15, pp. 116–132, 1985. G. S. Cheng and V. VanDorn, “Model-free adaptive control with cybocon,” in Techniques for Adaptive Control, New York: Elsevier Science, 2003, pp. 145-202. R. Badar and J. Shair, “Transient stability enhancement using sliding mode based NeuroFuzzy control for SSSC,” in Intelligent Systems and Applications, B. R. Bi Y., Kapoor S., Ed. Springer, 2017, pp. 3958.