552 JournalofElectricalEngineering

Published on June 2016 | Categories: Documents | Downloads: 91 | Comments: 0 | Views: 524
of 14
Download PDF   Embed   Report

Comments

Content

552 JournalofElectricalEngineering&TechnologyVol.5,No.4,pp.552~560,2010DOI: 10.5370/JEET.2010.5.4.552 Stability Analysis and Effect of CES on ANN Based AGC for Frequency Excursion J.Raja and C.ChristoberAsirRajan* Abstract This paper presents an application of layered Artificial Neural Network controller to study load frequency control problem in power system. The objective of control scheme guarantees that steady state error of frequencies and inadvertent interchange of tie-lines are maintained in a given tol- erance limitation. The proposed controller has been designed for a two-area interconnected power sys- tem. Only one artificial neural network controller (ANN), which controls the inputs of each area in the power system together, is considered. In this study, back propagation-through time algorithm is used as neural network learning rule. The performance of the power system is simulated by using conventional integral controller and ANN controller, separately. For the first time comparative study has been car- ried out between SMES and CES unit, all of the areas are included with SMES and CES unit sepa - rately. By comparing the results for both cases, the performance of ANN controller with CES unit is found to be better than conventional controllers with SMES, CES and ANN with SMES. Keywords: Artificial Neural Network, Automatic Generation Control, Capacitive Energy Storage, Stability Analysis, Superconducting Magnetic Energy Storage 1. Introduction Automatic Generation Control (AGC) is one of the most important issues in electric power system design, operation and control. The objective of the AGC in an interconnected power system is to maintain the frequency of each area and to keep tieline power close to the scheduled values by adjusting the MW outputs the AGC generators so as to ac- commodate fluctuating load demands. The AGC design with better performance has received considerable atten- tion during the past years and many control strategies have been developed for AGC problem. The controller based on classical control theories employed as in [1]-[4] are insuffi- cient because of changes in operating points while the loads change continuously during daily cycle, which is an inherent characteristics of power system. The variable structure controllers proposed as in [5], [6] are used to overcome this problem. On the other hand many adaptive is control techniques have been proposed for AGC [7]-[10]. Due to the requirement of the perfect model, which has to track the state variable and satisfy system constraints, it is rather difficult to apply these adaptive control techniques to AGC in practical implementations. These controllers are considerably slow because the estimation of parameter is needed and more computation is required. In this paper, ANN controller is used because this con - troller provides faster control than the others using convenCorresponding Author: Dept. of Electrical and Electronic Engineering, Sri ManakulaVinayagar Engineering College, Affiliated to Pondicherry University. Madagadipet 605 107, India. ([email protected]) * Dept. of Electrical and Electronic Engineering, Pondicherry Engineering College, Pondicherry, India.([email protected]) Received: April 23, 2010; Accepted: July 23, 2010 tional, adaptive control techniques. ANN controller gives a short settling time and eliminates the necessity of parame- ter estimation time required in conventional adaptive con- trol technique. The model of nonlinear controller which is the most suitable model to represent the real system is given by a set of differential equations, which is the most suitable model to represent the real system. As ANN con- figuration will be used to control the non linear system. So back propagation through time algorithm is preferred in the ANN controller to cope with the continuous time dynamics [11]-[13]. This algorithm in a way gives control rule. Even though multi layer perception is non recurrent, when more than one of this structure is used with back propagation through time algorithm, it can be perceived as a recurrent ANN. Thus this study, as the

system equations are used to model the system, there is no need to train ANN to obtain system model. It is well known that AGC system conventionally in- cludes an integral controller as secondary controller. The integral gain set to level the compromise between fast transient recovery and low overshoot in dynamic response of the system [14].Unfortunately, this type of controller even if its gain is optimized is considerably slow and therefore the recovery of transient in the power system with respect to load perturbation takes very long time. The considered power system includes two areas of the that consist of two steam turbine. The interconnected power system with two areas con- tains SMES units in each area. Addition of a small capacity SMES unit to the system sufficiently improves transients of frequency and tie line power deviation against to small load disturbances. In the SMES unit a small magnetic coil is connected to the ac grid through a power conversion unit, which includes converter and inverter. The super conduct -ing coil can be charged to set a value from the grid during normal operation of the power system. Once coil is charged, it conducts current with virtually no losses. When there is any sudden rise in demand, the stored energy is almost released through energy conversion device to the power system as an alternating current. The stored energy is taken about 20 MJ and 30 MJ because of factor of super conduc- tor stability. To damp out the oscillations as fast as possible, both the effect of SMES and CES based ANN using back propagation through time algorithm are investigated to- gether. For the simulation, step disturbance is included in both areas, the considered power system is controlled by using (i) Conventional integral controller. (ii) ANN using back propagation through time algorithm. The application of SMES to electric power system can be grouped into two categories. (i) Large scale energy stor- age like conventional pumped hydro plant storage to meet for sudden load leveling applications. (ii) Low capacity storage to improve the dynamic performance of power sys- tem. In first case large sized (hundreds of meters in diameter) high capacity super conducting magnetic energy stor- age capable of storing 108MJ is necessary [15]-[18], for the second application, very small sized SMES units with stor age capacity of the order of 102MJ or even less would be sufficient for a small load disturbances and with the opti- mized gain for the integral controller, the power frequency oscillation and the tie line deviation persist for a long dura- tion, the addition of a small capacity SMES unit to the sys- tem sufficiently improves this situation and the oscillations are practically damped out. The use of SMES and CES for load leveling application and for improvement of the dynamic performance of power system has been described in [18]-[21] The importance of control system using SMES has been presented as one of the powerful stabilizers for undamped oscillations which tends to occur in a long distance bulk power transmission system which has been viewed and analyzed in the litera- ture [22]. In [23] the improvement in AGC with the addi- tion of a small capacity SMES unit is studied, and time domain simulations used to study the performance of the power system dynamics are analyzed. Their applications in real power have invited problems form the view points of operation, maintenance, cost but the only advantages of SMES is, that is very much useful for high power applications. The battery energy storage for power system dynamic performance improvement has been widely and vividly reported as in [24], [25]. Their applications in real power system have invited problems from the view points of op- eration, maintenance and cost involvement. However, ca- pacitive energy storage (CES) [26] may be a better alternative choice to damp out the power frequency oscillation, following any perturbation in the power system. CES is, practically, maintenance free. Unlike magnetic energy storage units, CES does not impose any environmental prob- lem. The operation is quite simple and less expensive, compared to SMES. SMES requires a continuously operat- ing

liquid helium system. In magnetic storage systems, continuous flow of current is required but CES does not demand so. Thus, as a corrective measure against power system perturbation, CES plays an important role to damp out local mode power system oscillations. The objectives of the proposed works are: (i) to obtain the transient performance of the coordinated action of two areas AGC loop with various load conditions. (ii) To investigate the further impact of SMES on the same transient performance, control system using SMES has been pre- sented as one of the powerful stabilizers for damping out the power system oscillations. (iii) To investigate the im- pact of CES on the same transient performance, Control system using CES has been presented as one of the power- ful stabilizers for damping out the power system oscilla- tions. (iv) To compare the coordinated action of the tran- sient performance of AGC and ANN with SMES and without SMES. (v) To compare the coordinated action of the transient performance of AGC and ANN with CES and without CES. (v i) Finally comparative analysis has been carried between AGC and ANN with coordinated action of SMES and CES. The organization of the paper documented in the follow- ing headings. Section 2 provides two area power system investigations. While section 3&4 dealing with mathemati- cal modeling of SMES and CES. Section 5 incorporates ANN in AGC. Section 6 including Stability analysis com- parison with PI, SMES, CES. Section 7 Deals with result analysis being followed by concluding remarks. 2. AGC In Two-Area Power System The two area power system, including two single areas connected through a tie-line is considered. Each area sup- plies its user pool, and tie-line allows electric power to flow between areas. So, both areas affect each other (i.e., a load perturbation in one area affects the output frequencies of both areas as well as power flow on the tie-line). Because of this, the control system of each area needs infor- mation about the transient situation in both areas to bring the local frequency to its steady state value. As the information about each area is found in its output frequency, the information about the other area is in the perturbation of the tie-line power. So, tie-line power flow is needed in order to feed back the information in both areas. While the electric load increases in one area, the frequency of the same area decreases, and power transmitted from the other area to this area increases. In conventional systems, the turbine reference power of each area is tried to be set to its nominal value by an integral controller. The input of the integral controller of each area is Bi fi + Ptie (i = 1, 2), and it is called area control error (ACE). The parameters of Bi may be optimized, but here, they are chosen as 1/Kpi + 1/Ri as generally taken. The equations are: F(k1)f (k)(T /T )*(K *P (k)-K *P 1 1sp1p1t1p1e1 f (k)-K *P (k)) (1) 1 p1 12 P (k)(T /T )*(K *P (k)-P (k)) (2) t1 s t1 t1 h1 t1 J.Raja and C.ChristoberAsirRajan 553 554 Stability Analysis and Effect of CES on ANN Based AGC for Frequency Excursion P (k1)P (k)(T /T )*(K *P (k)-(K /R ) t2 h1 s h1 h1 ref1 h1 1 *f1(k)-Ph1(k)) (3) ft2(k1)P2(k)(Ts /Tp1)*(Kp1*Pt2(k) -(Kp1*Pe2) f2(k)Kp1 *P21(k)) (4) P (k1)P (k)(T /T )*(K *P (k) -P (k))(5) of SMES unit is intended to be settling its steady state value. If the load demand changes suddenly, the feed back provide the prompt restoration of current. The inductor current must be restored to its nominal value quickly after system disturbances, so that it can respond to the next load disturbance immediately. As a result, the equations of in ductor voltage deviation and current deviation for each area in Laplace domain are as follows: E (s)K di t2 t2st2t2h2t2 P (k1)P (k)(T /T )*(K *P (k) h2 h2s h2h2ref2

(Kh2/R2)*f2(k)Ph2(k)) (6) P (k1)2*pi*T *T *(f1(k)*f (k))P (k) (7) 11 f(s)K I (s) (12) 12 12s 2 12 oi 1ST i dci Idi 1ST di dci P (k1)2*pi*T *T *(f (k)-f (k))P (k) (8) 12 12s2121 P (k1)P (k)-(K *T )*(P (k)B *f (k))(9) ref1 ref1 i1 s 12 11 1 di SLdi Pref2(k1)Pref2(k)-(Ki2 *Ts)*(P21(k)B2 *f2) 3. SMES System (10) i change in increasing current, converter time delay, Koi (KV/Unit) is gain constant and Li (H) is the inductance of the coil. The deviation in the induc tor real power of SMES unit is expressed in time domain as follows. Psmi(t)diIdioIdidi (14) This value is assumed positive for transfer from AC grid to DC. The energy stored in SMES at any instant in time domain is given as follows. L I2 Wsmi(t) i di (MJ) (15) 2 3.1 Frequency Deviation as a Control Signal The frequency deviation f of the power system is sensed and used to control the SMES voltage, Ed. When power is to be pumped back in to the grid in the case of fall in the frequency due to sudden loading in the armature, the control voltage Ed is to be negative since the current through the inductor and thyristor cannot change its direc tion. The incremental change in the voltage applied to the inductor is expressed as: E  Kf*f (16) d  (1  S T ) Fig. 1 shows the transfer function model SMES unit contained DC Superconducting coil and converter which are connected by Star-Delta/Delta-Star transformer. The control of the converter firing angle provides the DC volt- age Ed appearing across the inductor to be continuously varying within a certain range of positive and negative val- ues. The inductor is initially charged to it s rated value Ido by applying a small positive voltage. Once the current reaches the rated value, it is maintained constant by reduc- ing the voltage across the inductor to zero since the coil is [24-25] superconducting. Neglecting the transformer and the converter losses, the DC voltage is given by equation (11) where Ed is DC voltage applied to the inductor (KV),  is firing angle (  ), Ed 2VdoCos2IdRc (11) Id is the current flowing through the inductor (KA).R is the equivalent commutating resistance (ohm) and Vdo is maximum circuit bridge voltage (KV). Charge and discharge of SMES unit are controlled through change of commutation angle. If is less than 90  , converter acts in rectifier mode and if is greater than 90, the converter acts in inverter mode. In this study, as in recent literature, inductor voltage deviation of SMES unit of each area is based on error of the same area in power system. Moreover the inductor current deviation is used as a negative feed back signal in SMES control Loop. So the current variable Fig. 1. Transfer Function Model of SMES. dc I (s) Where, Kidi is the gain for feedback. Idi ,Tdci is the Where: i,j=1,2, Ed is the incremental change in con - verter voltage, Tdc is the converter time delay, Kf is the gain of the control loop and S is the Laplace operator d/dt.If ACE is used as control signal then. E= Kf*{K( f+1 P)-K I} d(1+ST ) Ai 3.2 Area Control Error (ACE) As Control Signal E (s)

(13) iB ij IAi di (17) dc i In case where tie line power deviation signals are avail- able, it may be desirable to use area control error as input to SMES control logic. This has certain disadvantages, which are desirable later, compared to frequency deviation derived controls. The area control error of two areas is de- fined as: ACEi P + B *f (18) ijii Where, i,j=1,2, fi = Change in frequency of area, i. Pij = Change in tie line power flow out of area i to j. If ACE is directly used for the control of SMES, The gains constant KA (KV/unit Area) would be totally different from KF, the gain constant for frequency deviation as control signal. So a signal proportional to area control error ( fi + (1/Bi)* Pij) is used in such a scheme. Then, Edi  KAi (fi+ 1 *Pij) (19) (1  S T d c i ) Bi The discrete time equations for the two area AGC in- cluding SMES is solved in MATLAB using m-file and the results are presented here. The sampling time Ts is 0.0001.The equations of SMES for two area in discrete form are: E (k1)E (k)(T/T )*(B *f P ) di di dc ii tie *Koi Idi *Kid Edi(k)) (20) Idi (k 1)  Idi (k 1)  Idio (21) P (k1)E (k1)*(I (k1)I ) (22) smi di di dio 4. Capacitive Energy Storage Capacitor is an electro chemical device consisting of two porous electrodes, an ion exchange membrane separating the two electrodes and a potassium hydroxide electrolyte. In many ways, an ultra capacitor is subject of same physics as a standard capacitor. That is the capacitor is determined by [24]-[26] the effective area of the plates, the separation distances of the electrode and the dielectric constant of the separating medium. Fig. 2 shows transfer function model of CES. The opera- tion of CES units, that is, charging, discharging, the steady Fig. 2. Transfer function model of CES. state model and the power modulation during dynamic oscillatory period, is controlled by the application of the proper voltage to the capacitor so that the dsired current flows in to or out of the CES. This can be achieved by con - trolling the firing angle of the converter bridges. Neglect- ing the transformer and the converter losses, the DC voltage is given by [22] Ed 2VdoCos2IdRc (23) Where: Ed=DC voltage applied to the capacitor (Kv), =firing angle(degree), Id=current through the capaci- tor(KA),Rc=equivalent commutating resistance(ohm), Vdo=maximum open circuit bridge voltage of each 6-pulse converter at =0(Kv). The capacitor is initially charged to its normal volt- age,Edo by the PCS.Once the voltage of the capacitor has reduced Edo, it is kept floating at this voltage by continu - ing supply from the PCS to compensate for the dielectric and other leakage losses of the capacitor. The energy stored at any instant, J.Raja and C.ChristoberAsirRajan 555 CE 2 Wr d (MJ) (24) 2 Where C=Capacitance of CES (Farad). 4.1 Frequency Deviation as Control Signal The frequency deviation AF of the power system is sensed and used to control the CES current Id.Theincre- mental change in CES current is expressed as I di  Kcfi *f  (1  S T ) (25)

dci Where, i= 1,2. Where Idi is the incremental change in current of CES unit(KA),Tdci is the converter time delay(second),Kcfi is the gain of the control loop(KA/Hz),S is the Laplace opera- tor(d/dt) and I denotes the area. 4.2 Area Control Error (ACE) as Control Signal In case of the tie line power deviation signals are avail- able; it may be desirable to use area control error as input to CES control logic. It has certain advantages, compared to frequency deviation as control signal, which are de- scribed later. The area control error of two areas are de- fined as ACEi P + B *f (26) ijii Where, i= 1,2, fi = Change in frequency of area, i. Pij = Change in tie line power flow out of area i to j. If ACE is 556 Stability Analysis and Effect of CES on ANN Based AGC for Frequency Excursion directly used for the control of CES, The gains constant Kci (KA/unit Area) would be totally different from Kcf,the gain constants for frequency as control signal. So a signal pro- portional to area control error ( f + (1/B )* P ) is used in wlji (n) is the synaptic weight of neuron j in layer l that is fed from neuron I in layer l-1 such a scheme. Then, Idi Kcai (fi+ 1 *Pij) (27) Bi wherei,j=1,2. The capacitor voltage deviation can be sensed and used as a negative feed back signal in the CES control loop to achivev quick restoration of voltage then, that with fre- quency deviation as control signal, I 1 (Kf-kE) di (1STdci) cfiivdi di (28) Where Kvdi (KA/KV) is the gain corresponding to the Ed feedback. The block diagram representation of such a control scheme is shown in Fig. 3. 5. Neural Network Back-propagation Algorithm: The sequential updating of weights is the preferred method for on-line implementation of the back-propagation algorithm. For this mode of opera- tion, the algorithm cycles through the training sample {X(n),d(n)}nN1 as follows: (i) Initialization: Assuming that no prior information is available, pick the synaptic weights and thresholds from a uniform distribution whose mean is zero and whose variance is chosen to make the standard deviation of the induced local fields of the neurons lie at the transition between the linear and saturated parts of the sigmoid ac- tivation function. (ii) Presentations of Training Examples: Present the net- work with an epoch of training examples. For each ex- ample in the set, ordered in some fashion, perform the sequence of forward and backward computations de- scribed under points(iii) and (iv), respectively. (iii) Forward Computation: Let a training example in the epoch be denoted by (x(n), d(n)), with the input vector x(n) applied to the input layer of sensory nodes and the desired response vector d(n) presented to the output layer of computation nodes. Compute the induced lo- cal fields and function signals of the network by proceeding forward through the network, layer by layer. The induced local field V1j for neuron j in layer l is Assuming the use of sigmoid function, the output signal of neuron j in layer l is ylj(vj(n)) (30) j (1  S T d c i ) l mo l i1 vj(n)  wji(n)yi (n) (29) ej (n)  dj (n)  oj (n) (33) Where dj((n) is the jth element of the desired response vec- tor. (iv) Backward Computation: Compute the s (local gradients) of the network, defined by

l (n)  el (n) (vl (n)) (34) jjjj l l  l1 (l1) j (n)  j (vj (n)) k (n)wkj (n) (35) k Where the prime in j(,) denotes differentiation with re- spect to the argument. Adjust the synaptic weights of the Network in layer l according to the generalized delta rule: wl (n1)wl (n1)wl (n1)l(n)yi1(n) jijijiji(36) Where is the learning rate parameter and is the momen- tum constant. Iteration: Iterate the forward and backward computations under points (iii) and (iv) by presenting new epoch of train- ing examples to the network until the stopping criterion is met. The order of training examples should be randomized from epoch to epoch. The momentum and learning-rate parameter are typically adjusted (and usually decreased) as the number of training iterations increase. There are three states for the single area AGC. f - change in frequency of the plant output, PT - change in turbine power , PH change in hydraulic amplifier power. i0 Where: y i1 (n) is the output signal of neuron iin the i previous layer l-1 at iteration n. wl (n)bl(n) isthebiasappliedtoneuronjinlayerliiijjij if neuron j is in the first hidden layer (i.e. l=1), set yo x j(n) (31) j Where Xj(n)is the jth element of the input vector x(n).If neuronj is in the output layer, set ylo(n) (32) jj Compute the error signal J.Raja and C.ChristoberAsirRajan 557 Neural Network Controller: Fig. 3. Architecture of ANN controller for single area AGC. Hence, XP = [ f PT PH ] At steady state, f = 0.0, PT = PE, and PH = PE /KT1 The internal architecture of the neural network controller is shown in the above Fig. 3. The neural network consists of one input layer, one hidden layer of 20 neurons, and one output layer. In case of controller for single area system, we have one neuron at the output layer. In case of two area system, we use two neurons in the output layer for getting separate control inputs to the individual areas. In the input layer, we have got four nodes (in case of single area system). Three nodes are for the state inputs:change in frequency ( f), change in hydraulic power ( Ph), change in turbine power ( Pt) and fourth node for receiving the load estimate as input. Note that all the values are in per unit (P.U). The output neuron generates the control input to the plant (Uc). The activation functions used are hyper- bolic tangent functions for each neuron. Figure 3 is essen- tially composed of a layered arrangement of controller and plant equations blocks. The controller is a neural network. If the block containing the plant equations, P, were re- placed by a neural network copy, the unraveled system of Fig. 3 would become a giant layered neural network with inputs Xp(0) and PE, output Xp(K), and desired output Xd(K). The back propagation algorithm could then be ap- plied to train such a network. By doing so, the error gradi- ent defined at the output of the network are backpropagated through the the neural network copy of the plant and C blocks, from Xp(K) back to Xp(0); hence, the name back propagation through-time. This approach was first introduced by Nguyen and Windrow (1989), and was successfully applied to number of applications in the area of nonlinear neural control. In this paper, we adopt a slightly different approach by which we avoid the introduc- tion and training of a neural network copy of the plant equations. The basic idea is that instead of building a neu - ral network copy or emulator of P to back - propagate error gradients through it, it is possible to directly back propa- gate the error gradients through the plant equations P. Building a neural network emulator of the plant and back propagating error gradients through it is nothing other than approximating the true Jacobian matrix of the plant using a neural network technique. Whenever the equations of the plant are known beforehand, they can be used to

compute, analytically or numerically, the elements of the Jacobian matrix. The error gradient at the input of the plant is then obtained by multiplying the output error gradient by the Jacobian matrix. The inputs for ANN in two area AGC are the states of the plant. XP = [ f1, PT1, PH1, f2, PT2, PH2, P12 ] PE1 = PE1 PE2 = PE2 Where, PE1 is the load perturbation in area 1, PE2 is the load perturbation in area 2.For Training the vector XP was initially set to, XP = [0.0 0.0 0.0 0.0 0.0 0.0 0.0] The Target Output states of the plant during steady state are Xd=[ f1=0, PT1=( PE1+ P12), PH1= ( PE1+ P12)/KT1, f2=0, PT2=( PE2 - P12), PH2=( PE1- P12)/KT2, P12=0] 6. Stability of Control Loops The calculation of linearized system eigen values is done separately for the case where the SMES, or the CES con- trol is active. The calculated eigen values are presented in Table 1, 2 &3. Table 1 compares the calculated eigenval - ues of two area power system with and without controller. The calculated eigen value shows that for open loop, the system unstable because eigen values lies in right half of S plane and for the case of integral controller the system is stable. Table 2 shows that when the SMES loop is present system is stable and it reduces settling time, peak over shoot, similarly Table 3 compares the calculated eigenval- ues of power system with different cases, open loop, with SMES, with CES. The eigen values of Table 3 demon - strates the satisfactory behaviors of power system. For sta- bility analysis eigen values for two area open loop eigen values, with integral controller without SMES, with SMES, CES have been calculated separately. From this analysis AGC for two area with CES is better than AGC with SMES has been calculated. After this proposed controller CES and SMES has been compared properly. Table 1. Eigen Values S.N 1. PI controller -73.8646 Open Loop -72.5 -19.99 8.0031 0.0026 0.2405 0.4568 0.7958 2. -19.995 3. -3.9985 4. -0.0062 5. -1.6406 6. -0.5095 + 0.6308i 7. -0.5095-0.6308i 558 Stability Analysis and Effect of CES on ANN Based AGC for Frequency Excursion Table 2. Eigen values using SMES Table 5. Comparison table for Integral and ANN controller responses for Two Area power system Integral controller Response ANN controller Response S.N 1. 2. 3. 4. 5. 6. 7. 8. 9. Open Loop -72.5 -19.99 8.0031 0.0026 0.2405 0.4568 0.7958 - PI controller -73.8646 -19.995 -3.9985

With SMES -76.8646 -33.2545 -19.9985 Area 1 in% T f T Area 2 Area 1 Area 2 f T f T f -1.6406 -0.59 + 0.638i -0.59-0.638i - -1.6406 -0.5095 + 0.638i -0.5095 - 0.638i -0.0000 -0.0755 -1 -10 50 -50 Table 3. Eigen Values Using SEMS & CES controller, respectively. From this Table 4, 5, it is seen that the ANN controller is superior than conventional controller against to the step load disturbance increasing in the single and two areas. Further to damp out these oscillations AGC is imple- mented with SMES units in each area. These simulation results are carried out while SMES is present in each area of the power system, according to deviation of the power system energy demand, the SMES unit releases the needed energy or absorbs residue energy form power system. The model including generation unit and SMES unit together represents realistic performance of power system. So non linear state equation of the power system are used directly during the control of power system by ANN and integral controller. From the obtained results Table 6 & 7 shows that the performance of ANN controllers is better than con- ventional controller and also the positive effect of SMES unit on AGC is presented and SMES reduces settling and peak overshoot. Table 6. Comparative table for ANN based single areas AGC with/without SMES for various disturbances S.N 1. 2. 3. 4. 5. 6. 7. 8. 9. Open Loop -72.5 -19.99 8.0031 0.0026 0.2405 0.4568 0.7958 - With SMES -76.8646 -33.2545 -19.9985 -5.0062 -1.6406 -0.509 + 0.638i -0.509 - 0.638i 0.0000 -0.0755 With CES -82.8646 -37.2545 -33.9985 -25.0062 -21.6406 -5.5095 + 8.638i -5.5095 8.638i -0.2000 -2.0705 In this study, an application of ANN controller for load frequency control in power system is investigated. For the purpose, the interconnected power system having two areas consist of steam turbines is considered. First the load fre- quency control in power system is simulated by using both ANN and conventional controller to compare the behaviors of the controllers. In this simulation, a step load increasing in the first area of power system is considered, only one ANN controller, which controls the input of each area in power system together, is considered. Back propagation through time algorithm is used as neural network learning rule to cope with the continuous time dynamics. In Table 4, 5 shows frequency deviation in single, two area of power system against to the load disturbance mentioned above are shown for using ANN controller and conventional integral Table 4. Comparison table for integral and ANN controller for Single Area power system Load in % 1 10 -1 -10 50 -50 Without SMES With SMES Table 7. Comparative table for ANN based two areas AGC with/without SMES for various disturbances (1%, 10%, 50%, -1%, -10%, -50%) Load in % Integral controller ANN controller Area 1 T f (Hz) (s) 38 0.001 36 0.003 43 0.006 42 0.002 45 0.003 20 0.002 20 0.028

Area 2 T f (Hz) (s) 38 0.0001 31 0.0003 43 0.0009 42 .00002 45 .00003 20 .00002 20 .00028 Area 1 T f Hz T (s) (s) Area 2 f (Hz) Time (sec) 8.7 10.83 12.78 11.18 f (Hz) -0.000182 -0.000512 0.000016 0.000424 Time (sec) 2.4 3.5 2.8 2.5 1.7 4.1 f (Hz) -0.000182 -0.000512 -0.000091 -0.000468 -0.005073 0.009031 1 10 -1 -10 50 9.37 -0.006025 1.5 -0.0001 1.6 -0.0006 1.5 -0.003 1.5 0.0001 1.5 0.0006 1.0 -0.0001 1.0 -0.001 1.5 0.0001 1.5 0.0005 1.5 0.0029 1.5 0.0001 1.5 0.0005 1.0 0.0001 1.0 -0.001 -50 9.37 0.006025 7. Result Analysis Load -0.0062 -5.0062 10 (s) Hz 36 0.03 43 0.06 42 0.02 45 0.03 20 0.02 (s) Hz (s) 36 0.03 31 43 0.06 43 42 0.02 42 45 0.03 45 20 0.02 20 Hz 0.003 0.009 0.002 0.003 0.002 (s) Hz 31 0.0003 43 0.0009 42 0.0002 45 0.0003 20 0.0002 Time (sec) 2.4 3.5 2.8 2.5 1.7 4.1 f in Hz -0.000182 -0.000512 -0.000091 -0.000468 -0.005073 0.009031 Time (sec) 1.5 2.0 1.8 1.9 1.5 3.0 f (Hz) -0.000151 -0.000452 -0.000085 -0.000365 -0.003562 0.00731 Without SMES Including SMES J.Raja and C.ChristoberAsirRajan 559 The validity of superconducting magnetic energy storage and battery energy storage for power system dynamics per- formance improvement has been widely reported in [10][18]. Their applications in real power system have invited problems from the view point of operation, maintenance and cost involvement but capacitive energy storage (CES) is a better choice to damp out power system oscillations, following any perturbation in power system. CES is practi- cally, maintenance free, unlike other energy storage devices, and CES does not impose any environment problem. The operation is quite simple and less expensive. Thus a correc- tive measure against power system perturbation, CES plays an important role to damp out local modes of oscillations. Table 8 shows performance evaluation of CES with two areas is considered. From this table it is noted that due to the application of CES action for either 10% to 50% step disturbance. Under this condition, the stored energy in CES is almost immediately released through the power conver- sion system to the grid. Thus, application of capacitive energy stored unit quickness the transient stability phe- nomena even after the occurrence of a fault and subsequent clearance of the same and it may be successfully practi- cally implemented for improving small signal dynamic performance of power system. Table 8. Time Vs f for two areas AGC with SMES & CES for Various disturbances 1%, 10%, 50%, 75%, -1%, -10%,-50%, -75%) power system, in most cases. ANN with back propagation through time algorithm is used as controller and power system is modeled by its state space equations. In this

work the obtained simulation results show that the performance of ANN with CES is better than conventional controller, with energy storing devices and ANN with SMES controller. References [1] J. K. Cavin, M.C. Budge, and P.Rosmusen, An op - timal linear system approach to load frequency con- trol , IEEE Trans. On power apparatus and system, Vol.90, pp. 2472 2482, 1971. [2] N.N.Benjamin and W.C.Chang , Multilevel load fre- quency control of interconnected power system , Proc, IEE, Vol. 125, pp. 521 -526, 1978. [3] J.NandaabdB.L.Kavi, Automatic Generation con- trol of interconnected power system, Proc, IEE, Vol. 125, No.5, pp.385 -390, 1988. [4] D.Das, J.Nanda, M.L.Kothari, and D.P.Kothari, Auto - maticgenerationcontrolof hydro thermal system with new control error considering generation rate con- striaint, Electric machines and power system, Vol.18, pp. 461-471, 1990. [5] N.N.Benjamin and W.C.Chang, Variable structure control of electric power generation, IEEE Trana.on power Apparatus and System, Vol. 101, No.2, pp. 376 -380, 1982. [6] S. Haykin, Neural Networks: A Comprehensive Foundation 2nd edition, 2006. [7] D.H. Nguyen and B.Widrow, 1990, Neur al Net- works for Self learning Control system, IEEE Con- trol Systems Mag., pp.18-23, 1990. [8] D. H. Nguyen and B. Widrow, Neural networks for self-learning control systems, IEEE Control Systems Mag., pp. 18 23, 1990. [9] S. C. Tripathy, R. Balasubramania, and N. P. S. Chan - ramohanan, Effect of superconducting magnetic en- ergy storage on automatic generation control considering governor deadband and boiler dynamics, IEEE Trans. on Power System, Vol. 7, No. 3, pp. 1266-1273, 1992. [10] A. Demiroren, N. S. Sengor, and H. L. Zeynelgil, Automatic generation control by using ANN tech- nique, Electric Power Components and systems (Electric Machinesand Power Sytems), Vol. 29, No. 10, pp. 883 -896, 2001. [11] F.Beaufays, Y. A. Magid, and B. Widrow, 1994, Ap- plication of Neural Network to Load Frequency Con- trol in Power System, Neural Networks, Vol.7, No.1, pp.122 -128, 1989. [12] A. Demiroren, N.S.Sengor, and H.L.Zeynelgi, Auto- maticgenerationcontrolby using ANN technique, Electric Power Component and Systems.Vol.29, No.10.pp.883 -896,2001. [13] H.L.Zeynelgil, A. Demiroren, and N.S.Senger, Load Frequency control for power saystem with reheat With SMES With CES Area 1 Area 2 Area 2 Area 1 f (Hz) T (s) f ( Hz) T (s) f (Hz) T (s) -0.0001 1.5 0.0001 1.4 0.0005 1.4

-0.0006 1.5 0.0005 1.5 0.0004 1.5 -0.003 1.5 0.0029 1.3 0.0003 1.3 0.0001 1.5 0.0001 1.2 0.0003 1.2 0.0006 1.5 0.0005 1.2 0.0004 1.2 -0.0001 1.0 -0.0001 0.8 0.0003 0.8 -0.001 1.0 -0.001 0.8 0.0005 0.8 T f (s) 1.5 ( Hz) .0005 1.6 0.0004 1.5 0.0003 1.5 0.0003 1.5 0.0004 1.0 0.0003 1.0 0.0005 8. Conclusions This study is an application of ANN to automatic gen- eration control, in a power system, having SMES and CES. In this work, transient behaviors of the frequency of each area and tie line power deviation in the power system with two areas, are considered, under any load disturbances in any area. In practice, power system generally has more than two areas, and each area is different from others. But, in this study, the power system, of

two thermal areas, is considered. The nonlinear state space equation of power system is obtained and these equations are used directly during the control of power system by both integral con- troller and ANN controllers. This is not an usual method with ANN controller. When ANN used in order to back propagate the error, ANN emulator is used, instead of 560 Stability Analysis and Effect of CES on ANN Based AGC for Frequency Excursion steam turbine and governor dead band normality by using Neural Network Controller, European Tran- scation on Electrical power (ETEP), Vol.12, No. 3, pp.179 -185, 2002. [14] O.I.Elgerd, Electric Energy System Theory: An Intro- duction, McGraw Hill Book Company, 1971. [15] H.J.Kunish, K.G.Krammer and H.Domonic Battery Energy Storage-Another Option for Load Frequency Control & Instantaneous reserve , IEEE Trans on Energy conversion Vol. E-1, No. 3. pp41-46, Sep 1986. [16] H.J.Boeniig& J.F.Haur, Commisioning tests of the Bonneville power Administration 30 MJ super con- ducting Magnetic storage unit , IEEE Trans on power apparartus& systems.Vol. PAS-104, No. 2. PP.302 - 312. Feb 1982. [17] Banarjee.J.K, Chatterjee and S.C.Tripathy, Applica- tion of Magnetic energy storage units as Load Fre- quency Control Stabilizers , IEEE Trans, on energy conversion, Vol. 5, No. 1, pp.481-501, March 1990. [18] Kwa-Sur Tam and Premkumar, Application of super magnetic energy storage in an asynchronous link be- tween power system , IEEE Trans on Energy conver- sion, Vol. 5, No. 3, Sep 1971. [19] H.A. Peterterson, N.Mohan and R.W.Boom, Super Conductive energy storage Inductor-Converter Units for power system , IEE Trans. On Powerapparatus and systems, Vol. PAS-94, No. 4, PP 1337 -1348, July/August 1975. [20] R.J. Loyd, J.D.Rogers et al., A feasibility utility scale superconducting Magnetic Storage Plant , IEEE Trans. On Energy conversion, Vol. EC-1, No. 4, PP. 63 -64, December 1986. [21] Y.Mitani, K.Tsuji, Y.Murakami, Application of Su - perconducting Magnetic Energy storage to improve power system dynamic performance, IEEE trans. On power system, Vol. 3, No. 4-1988. [22] S.C. Tripathy, R.Balasubramaniam, P.S. Chandramo - han Nair, Effect of Superconducting Magnetic En- ergy Storage on Automatic Generation Control Considering Governor Dead band and Boiler dynamics. IEEE tran. On power system Vol 7,Aug 1992. [23] S.C. Tripathy, K.P. Juengst, Sampled data automatic generation control with superconducting magnetic energy storage in power systems, IEEE Trans. Energy Convers.12 (June (2)) (1997) 187 -192. [24] S. Banerjee, J.K. Chatterjee, S.C. Tripathy, Applica - tion of magnetic energy storage unit as load fre- quency stabilizer, IEEE Trans. Energy Convers. 5 (March (1)) (1990) 4651. J.Rajaborn on 1980 and received his B.E.degree (Electrical and Electronics) and M.E. degree (Power System) in the year 2001 & 2003 respectively ,He is currently pursuing his Ph.D degree in Pondicherry University, Puducherry, India. He published technical papers in International & National Journals and Conferences. He is currently working as Assistant Profes- sor in the Electrical & Electronics Engineering Department at Sri ManakulaVinayagar Engineering College, Affiliated to Pondicherry University, Pondicherry, India. His area of interest is power system Controls and Stability, operational planning and control. He acquired Member in ISTE in India. C.ChristoberAsirRajanborn on 1970 and received his B.E. (Distn.) degree (Electrical and Electronics) and M.E. (Distn.) degree (Power System) from the Madurai Kamaraj University (1991 & 1996), Madurai, India. And he received his postgraduate degree in DI.S. (Distn.) from the Annamalai

University, Chidambaram. He received his Ph.D degree in Power System from the College of Engineering, Guindy, Anna University, Chennai, India (2001 -2004). He published technical papers in International & National Journals and Conferences. He is currently working as Associate Professor in the Electrical & Electronics Engineering Department at Pondicherry Engineering College, Pondicherry, India. His area of interest is power system optimization, operational planning and control. He is undertaking various R & D projects. He acquired Member in ISTE and MIE in India. Appendix-1 System Data: Pr1=Pr2=1200MW, Tp1=Tp2=20s, Kp1=Kp2=120Hz/p.u.MW, Tr1=Tr2=10s, Kr1=Kr2=0.5, Tt1=Tt2=0.3s, T12=0.0866s, Tg1=Tg2=0.08s, R1=R2=2.4Hz/pu.MW, D1=D2=8.33*10^ -3 puMW /Hz, P1=0.01pu/MW, P2=0.0pu/MW. Super Conduction Magnetic Energy Storage Data: Kf=100KV/unit MW, Kid=0.20KV/KA, KAi=0.875, L=2.65H, Tdc=0.03s, Iido=4.5KA. Capacitive Energy Storage Data: C=1farad, R=100 ohm, Tdc=0.05s, Kace=70KA/unit MW, Kvd=0.1KA/KV, Vdo=2KV.

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close