You are on page 1of 6

Using a Combined Artificial Intelligent Approach In Distance Relay For Transmission Line Protection In EPS

T. S. Kamel M. A. Moustafa Hassan A. El-Morshedy


Elec. Power and Machines Department, Faculty of Engineering, Cairo University, Egypt.
Corresponding E-Mail: mmustafa@eng.cu.edu.eg AbstractThe sophisticated Application of Artificial Intelligent Approaches was introduced recently in Protection of Transmission line in electric power systems. However, these approaches started with introducing Fuzzy Logic (FL) in the last decades of the last century. Furthermore, Artificial Neural Network (ANN) was introduced to solve many problems in electric power systems. Among these problems is Protection of Transmission line. In this proposed article, the application of Adaptive Neuro-Fuzzy Inference System (ANFIS) for Distance Relay Protection for Transmission line in Electrical Power systems (EPS) will be illustrated. The proposed technique is focusing on fault detection, classification, and location. Furthermore, all these issues will be addressed. The ANFIS can be viewed as a fuzzy system, a neural network or fuzzy neural network. This paper aims; firstly, to detect the fault occurrence in very short time and isolate the faulty section of the transmission lines. Secondly to classify the fault type and deduce which of the three phases are exposed to the fault. Finally, locating the fault will be achieved easily. The input data of the ANFIS are firstly derived from the fundamental values of the voltage and current measurements (using digital signal processing via Fourier transform). KEYWORDS: Distance protection, Adaptive Neuro Fuzzy Inference system (ANFIS), Artificial Intelligent Approach, Fault Detection, Fault Classification, Fault location I. INTRODUCTION Distance protection relays are commonly used for the protection of transmission lines. Since the impedance of a transmission line is proportional to its length, such a relay is based on impedance calculation. Although the fundamentals of distance protection were started long time [1-2], theoretical foundations as well as practical applications of distance protection are still common topics of investigation. With the recent digital technology being ever increasingly and adopted in power systems, more particularly in the protection field, distance relays have experienced some improvements, mainly related to efficient filtering methods (such as Fourier, Wavelet, Kalman, etc.). As a result, shorter decision time will be achieved. The trip and no trip decision will be improved, compared to electromechanical or solid state relays. Furthermore, the conventional distance protection is usually designed on the basis of fixed conventional relay settings. The reach accuracy of a distance relay can therefore be affected by the different fault conditions as well as network configuration changes. Besides that these Schemes are deterministic computations assuming system modelling based on conventional mathematical tools. Such system representation is not well suited for dealing with ill-defined and uncertain systems. On the other hand, intelligent computational techniques such as Fuzzy Inference Systems (FIS), Artificial Neural Network ANN, and ANFIS can model qualitative aspects of human knowledge and restoring process without employing quantitative analysis. Thus these approaches are fetching great attention nowadays, especially with the absence of a simple and well-defined mathematical model. These models are characterized by non-random uncertainties associated with vagueness and imprecision in real-time systems [3-4]. Recently some interesting research efforts using neural network and fuzzy-logic based techniques for transmission line fault classification and location have been published. Such approaches involve removal of D.C offset and nonharmonic components as well as determination of sequence components of line currents [5-11]. This proposed research utilizes a combined Artificial Intelligent Approach. This approach utilizes the adaptive-network-based fuzzy inference system. The main advantage of this approach is to overcome the difficulties associated with conventional voltage and current based measurements for transmission line distance relaying algorithms. These difficulties are due to the effects of some factors such as fault inception angle (or fault inception time), fault impedance and fault distance. This paper is integrating the learning capabilities of neural network to the robustness of fuzzy logic systems. Neural network has the shortcoming of implicit knowledge representation, whereas, fuzzy logic systems are subjective and heuristic. The determination of fuzzy rules, input and output scaling factors and choice of membership functions depend on trial and error that makes the design of fuzzy logic system a time consuming task. These drawbacks of neural network and fuzzy logic systems are eliminated via the integration between the neural network technology and the fuzzy logic systems in one combined approach. It also provides a natural framework for combining both numerical information in the form of input/output pairs and linguistic information in the form of IFTHEN rules in a uniform fashion. The main items of this research will be discussed in the following sections, as well as some of the results will be presented at the end. II. BACKGROUND OF ANFIS

9781-4244-3428-2/09/$25.00 2009 IEEE

In general, a Fuzzy Logic System (FLS) can be viewed as a non-linear mapping from the input space to the output space. An FLS consists of five main components: fuzzy sets, fuzzifiers, fuzzy rules, an inference engine, and defuzzifiers [17]. However, Fuzzy inference system is limited in its application to only modelling ill defined systems. These systems have rule structure which is essentially predetermined by the user's interpretation of the characteristics of the variables in the model. It has been considered only fixed membership functions that were chosen arbitrarily. However, in some modelling situations, it cannot be discerned what the membership functions should look like simply from looking at data. Rather than choosing the parameters associated with a given membership function arbitrarily, these parameters could be chosen so as to tailor the membership functions to the input/output data in order to account for these types of variations in the data values. In such case the necessity of the Adaptive Neuro Fuzzy Inference System (ANFIS) becomes obvious. Adaptive Neuro-fuzzy networks are enhanced FLSs with learning, generalization, and adaptive capabilities. These networks encode the fuzzy if-then rules into a neural network-like structure and then use appropriate learning algorithms to minimize the output error based on the training/validation data sets. Neuro-adaptive learning techniques provide a method for the fuzzy modelling procedure to learn information about a data set. It computes the membership function parameters that best allow the associated fuzzy inference system to track the given input/output data. A network-type structure similar to that of a Artificial Neural Network (ANN) can be used to interpret the input/output map. Therefore, it maps inputs through input membership functions and associated parameters, and then through output membership functions and associated parameters to outputs,. The parameters associated with the membership functions changes through the learning process. The computation of these parameters (or their adjustment) is facilitated by a gradient vector. This gradient vector provides a measure of how well the fuzzy inference system is modelling the input/output data for a given set of parameters. When the gradient vector is obtained, any of several optimization routines can be applied in order to adjust the parameters to reduce some error criteria. This error criterion is usually defined by the sum of the squared difference between actual and desired outputs. ANFIS in the MATLAB program [13] uses a combination of least squares estimation and back propagation for membership function parameter estimation. Furthermore the used ANFIS is assumed to have the following properties: It is zero th order Sugeno-type system. It has a single output, obtained using weighted average defuzzification. All output membership functions are constant. It has no rule sharing. Different rules do not share the same output membership function, namely the number of output membership functions must be equal to the number of rules. It has unity weight for each rule.

Figure (1) shows the architecture of the ANFIS, comprising by input, fuzzification, inference and defuzzification layers. The network can be visualized as consisting of inputs, with N neurons in the input layer and F input membership functions for each input, with F*N neurons in the fuzzification layer. There are F^N rules with F^N neurons in the inference and defuzzification layers. It is assumed one neuron in the output layer. For simplicity, it is assumed that the fuzzy inference system under consideration has two inputs x and y and one output z as shown in Figure (1). For a zero-order Sugeno fuzzy model, a common rule set with two fuzzy if-then rules is the following: (1) Rule 1: If x is A1 and y is B1, Then f1=r1 (2) Rule 2: If x is A2 and y is B2, Then f2 = r2 Here the output of the ith node in layer n is denoted as On,i: Layer 1 Every node i in this layer is an adaptive node with a node function: for i=1,2,3 or (3) O1,i=Ai(x) for i=4,5,6 (4) O1,i =Bi-3(y) Where x (or y) is the input to node i and Ai (or Bi) is a linguistic label associated with this node. In other words, O1,i is the membership grade of a fuzzy set A1, A2 and A3 (or B1, B2 and B3) and it specifies the degree to which the given input x (or y) satisfies the quantifier A (or B). Here the membership function for A (or B) is triangular membership function and is given as: 1 Left: (u)= Max 0, 1 + cL-u 0.5wL 0, 1+ c-u / 0.5w 0, 1 + u-c / 0.5w
L

if u cL (5) otherwise

Max Centers: C(u)= Max 1 Right: R(u)= Max

otherwise (6) if u c otherwise (7)

0, 1 + ( u- c )/ 0.5w

if u cR

Notice that for Equation (5) cL specifies the saturation point and wL specifies the slope of the nonunity and nonzero part of L as shown in Figure (2) Similarly, for R. For C notice that c is the centre of the triangle and w is the base-width. cL, cR, c, wL, wR, and w are the parameters set. As the values of these parameters change, the triangular function varies accordingly, thus exhibiting various forms of membership functions for fuzzy set A. Parameters in this layer are referred to as premise parameters. Input Inputmf Rules Outputmf Output

c1=

Figure (1): The Architecture of the ANFIS Layer 2 Every node in this layer is a fixed node whose output is the product of all the incoming signals: (8) O2,i= w i= Ai (x) Bi (y) i=1,2,3 Each node output represents the firing strength of a rule. Layer 3 Every node i in this layer is an adaptive node with a node function: i=1,2,3 (9) O3,i = wi fi = wi ri Where ri is the parameter set of this node. Parameters in this layer are referred to as consequent parameters.

This is linear in the consequent parameters r1, r2, and r3. From this observation, it can be concluded that: S = set of total parameters, S1 = set of premise (nonlinear) parameters, S2 = set of consequent (linear) parameters Therefore the overall output will be: (14) O4,i = F(i, S) Where i is the vector of input variables, F is the overall function implemented by the adaptive network, and S is the set of all parameters which can be divided into two sets: S = S1 S2 (15) Where represents direct sum. Therefore, the hybrid learning algorithm can be applied directly. More specifically, the error signals propagate backward and the premise parameters are updated by Gradient Descent (GD) and node outputs go forward until layer 3 and the consequent parameters are identified by the Least Squares (LS) method. This hybrid learning is organized as follows: a) Linear and nonlinear parameters are distinguished b) Each iteration (epoch) of GD update the nonlinear parameters c) LS follows to identify the linear parameters. III. SIMULATION SCHEME

w1 w2 , c2= , w1 + w2 + w3 w1 + w2 + w3 w3 c3= w1 + w2 + w3

(13)

Figure (2) : Input membership functions (triangular) Layer 4 The single node in this layer is a fixed node which computes the overall output as the summation of all incoming signals: Overall output = O4,i

wi. fi = wi
i i

i=1,2,3

(10)

The simulation environment is based on MATLAB software package. This package is selected as the main engineering tool for performing modelling and simulation of power systems and relays. ATP program [14] is used for detailed modelling of power network and simulation of interesting events. Scenario setting and relaying algorithm will be implemented in MATLAB program while the data generation for training and testing this algorithm will be executed by ATP program. IV. THE T. LINE UMDER PROTECTION A single line diagram for the protected transmission line T.L is shown in Figure (3). It is consists of two circuits of 80 km length, 66 kV voltage level and 2 GVA short circuit level.

From the ANFIS architecture shown in Figure (1), it is noticed that when the values of the premise parameters are fixed, the overall output can be expressed as a linear combination of the consequent parameters. In symbols, the final output in Layer 4 can be rewritten as: O4,i

wi. fi = = wi
i i

As w1, w2 and w3 are assumed to be constant. Therefore, equation (11) can be rewritten as follows: (12) O4,i =c1.r1+c2.r2+c3.r3 Where

w1 w2 w 3 f 1+ f 2+ f3 w1 + w2 + w3 w 1+ w2 + w3 w 1+w2+w 3

(11) Figure (3): Single line diagram for the T.L V. PROPOSED PROTECTION SCHEME At the instance of the fault in the transmission line the Fault Detection unit identifies the occurrence of the fault. Then it

activates the Fault Classification and Fault location units. The Fault Classification unit will interpret the faults values of the voltage and the current signals. According to these values, it will be capable of determining the fault type if it is single line to ground, double lines, double lines to ground, or three phases. On the other hand the Fault location unit will also take the faulty values of the voltage and the current signals. According to these values, it will be able to estimate the location of the fault from the relaying point. Finally the control unit will receive the decisions from the three units in which the output of the fault detection unit will trip the circuit breaker at the fault instance through the control circuit as shown in Figure (4). However, the outputs of the Fault classification and the Fault location units will offer an aid to identify the fault type and location respectively. Figure (4) illustrates the layout of the proposed distance relay based on the combined intelligent approach. The proposed distance relay consists mainly from four main units: - Fault Detection Unit - Fault Classification Unit - Fault locator Unit - Control Unit Fault Detection Unit represents the main unit in the proposed transmission system protection scheme.

The ANFIS detector consists of six neurons in the input layer i.e. N=6, three triangular membership functions for each input i.e. F=3 and constant membership function for the output as described in [15 & 16]. iii) Testing data for Fault Detection Unit: The testing data are chosen at different fault and no fault conditions. The fault conditions are done at different fault distances, different fault resistances, different fault inception times and different fault types which are not chosen among the training data. Some of them are shown in Table 1. The no fault conditions are performed at different states of variations in the voltage and the frequency of the two generators within the allowable limits to emulate the deviations in the feeding and the loading conditions in the power systems. Some of them are shown in Table 2 where the variations in the voltage and the frequency of the two generators are introduced. a) Fault Classification Unit:

Fault Classification Unit is the next unit to be activated from the Fault Detection Unit in case of fault, as presented in Figure (5). i) Training data for Fault Classification Unit: The training data used to train the ANFIS of the fault classification unit are taken at fault distance (Df) 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70% and 80% of the line at all type of faults (i.e. single phase to ground, phase to phase, double phase to ground or three-phase fault) with inception fault time (Tf) 2 msec. and fault resistances (Rf) 0, 25, 50 and 100 ohms. There are 396 training data .The input data to the ANFIS Classification are the impedances of the three phases (magnitude and phase) and the zero sequence component of the currents (i.e. 7 input) after divided them by their non fault values. They are taken from the fundamental values of the voltage and current measurements after making Fourier transform every 20 msec. While the output data from ANFIS are: 0.5output<1.5 for single phase to ground fault 1.5 output< 2.5 for phase to phase fault 2.5output<3.5 for double phase to ground fault 3.5 output< 4.5 for three-phase fault ii) The ANFIS classifier: The ANFIS classifier consists of seven neurons in the input layer i.e. N=7, three triangular membership functions for each input i.e. F=3 and constant membership function for the output as described in [15] iii) Testing data for Fault Classification Unit: The testing data are chosen at different fault conditions. These data are carried out at different fault distances, different fault resistances, different fault inception times and different fault types. However, these data are not among the training data. Some of results with these testing data are shown in Table 3. b) Fault Location Unit: Fault Location Unit process the results from the detection unit. Classification unit and Fault Location Unit are working in parallel as illustrated in Figure (5).

Figure (4): The Layout t of the proposed Distance relay i) Training data for Fault Detection Unit: The used training data to train the ANFIS of the fault detection unit are taken at the no fault conditions and fault conditions. The fault conditions are carried out at all different fault types at fault distance (Df) 5%, 40% and 80% of the line with inception fault time (Tf) 5 msec. and fault resistances (Rf) 0 and 100 ohms. There are 69 training data. The input data to the FNN detection unit are the impedances of the three phases (magnitude and phase i.e. 6 input) after divided them by their non fault values. They are taken from the fundamental values of the voltage and current measurements after making Fourier transform every 10 msec. While the output data from ANFIS are: -1output<0.5 for no fault conditions 0.5 output< 3 for fault conditions ii) The ANFIS detector:

output number which represents the fault types for the fault location task, all the testing data for the ANFIS locator gives a maximum error 8% for different fault resistances, different fault inception times and different fault types. REFERENCES [1] [2] [3] [4] Figure (5) : The Detailed Protection Scheme [5] i) Training data Fault Location Unit: The training data used to train the ANFIS of the fault location unit are taken at fault distance (Df) 5%, 10%, 15%, 20%, 30%, 40%, 50%, 60%, 70% and 80% at all type of faults (i.e. single phase to ground, phase to phase, double phase to ground or three-phase fault) with inception fault time (Tf) 2 msec. and fault resistances (Rf) 0, 25, 50 and 100 ohms. There are 444 training data. The input data to the ANFIS of the location unit are as same as those of the ANFIS of the classification unit without the zero sequence current. The output data from the ANFIS are the normalized fault distance value. ii) The ANFIS locator The ANFIS locator consists of six neurons in the input layer i.e. N=6, four triangular membership functions for each input i.e. F=4 and constant membership function for the output. iii) Testing data Fault Location Unit: The testing data are chosen randomly with different fault conditions which are carried out at different fault distances, different fault resistances, different fault inception times and different fault types. However, these set are not chosen before for the training data. Besides that a white noise is introduced in the testing data to model the errors in the voltage and current measurements. Some of the results with these testing data are shown in Table 4. CONCLUISON An Artificial Intelligent Approach for Distance Relay in Transmission line Protection was introduced in this paper. The proposed Artificial Intelligent Approach for distance relay demonstrates successful performance for the three main protection tasks. For the fault detection task, all the testing data for the ANFIS detector in the fault and no fault conditions give the correct output within the given periods that mentioned before. For the fault classification task, all the testing data for the ANFIS classifier gives the correct [12] [13] [14] Sunil S. Rao, "Switchgear and Protection", 10th Edition, KHANNA Publishers, Delhi, 1994. W. Mark Carpenter "IEEE Guide for Protective Relay Applications to Transmission Lines", IEEE Std C37.113, 1999. Jacek M. Zurada , "Introduction to Artificial Neural Systems",1st Edition, PWS Publishing Company, Boston, 1995. Kevin M. Passino and Stephen Yurkovich, "Fuzzy Control", 1st Edition, Addison Wesley Longman, Inc., California, 1998. Dalia Farouk Mohamed," A New Design of an Intelligent Digital Distance Protective Relay" PhD Dissertation Submitted to the Office of Graduate Studies of Cairo University, 2007. SlavkoVasilic," Fuzzy Neural Network Pattern Recognition Algorithms For Classification Of The events In Power System Network", PhD Dissertation Submitted to the Office of Graduate Studies of Texas A&M University 2004. Abeer Galal Saad ,"Digital Relaying of High Voltage Transmission Lines by Artificial Neural Networks", Master Dissertation Submitted to the Office of Graduate Studies of Cairo University 2004. P. K. Dash, A. K. Pradhan, and G. Panda: "A Novel Fuzzy Neural Network Based Distance Relaying Scheme", IEEE Transactions on Power Delivery, Vol. 15, No. 3, pp.902-907, 2000 D. V. Coury and D. C. Jorge, "Artificial Neural Network Approach to Distance Protection" IEEE Transactions on Power Delivery, Vol. 13, No. 1, pp. 102-108, 1998. Huisheng Wang and W. W. Keerthipala, "FuzzyNeuro Approach to Fault Classification for Transmission Line Protection" IEEE Transactions on Power Delivery, Vol. 13, No. 4, pp.1093-1104, 1998. M. Jayabharata Reddy and D.K. Mohanata, "Performance Evaluation of Adaptive Network Based Fuzzy Inference System Approach for Location of Faults on Transmission Lines Using Monte Carlo Simulation" This paper has been accepted for publication in a future issue of IEEE journal, but has not been fully edited, 2007. MATLAB R2008a. ATP Draw version 3.5 Tamer S. Kamel, and M. A. Moustafa Hassan, OnLine Trained Adaptive Neuro-Fuzzy Inference System for Distance Relay of Transmission line Protection, presented at The World Congress on Electronics and Electrical Engineering, (WCEEENG'09), April 6-9, 2009, Cairo, Egypt.

[6]

[7]

[8]

[9]

[10]

[11]

[15]

Tamer S. Kamel, and M. A. Moustafa Hassan, [16] H. Khorashadi Zadeh and Zuyi Li, Transmission Adaptive Neuro Fuzzy Inference System For Fault Line Distance Protection Using ANFIS and Positive Location in Transmission Lines, presented at The Sequence Components, Proceedings of the iREP World Congress on Electronics and Electrical Symposium- Bulk Power System Dynamics and Engineering, (WCEEENG'09), April 6-9, 2009, Cairo, Control - VII, Revitalizing Operational Reliability Egypt. August 19-24, 2007, Charleston, SC, USA Table 1 Testing data of the Fault Detection Unit and their Outputs for the fault conditions Tf 0.007 0.007 0.002 0.002 0.004 0.004 Rf 50 50 100 100 25 25 Df p.u 0.1 0.1 0.6 0.6 0.35 0.35 Fault type B-G B-C A-B-G A-B-C C-G A-B-C-G Za p.u 0.95 0.97 0.2 0.27 1.02 0.17 Za ph 160.2 157.5 124.4 49.1 157.9 57.3 Zb p.u 0.45 0.49 0.99 0.13 0.94 0.12 Zb ph -265 -270 -244 -325 -245 -320 Zc p.u 1.06 0.57 0.2 0.23 0.55 0.19 Zc ph 156.6 164 98.9 108.9 120.7 100.2 Output 1.8 2.4 1.3 1 1.3 1

Table 2: Testing data of the Fault Detection Unit and their Outputs for the no fault conditions % V1 -5 -5 0 5 % V2 0 -5 5 5 f1 Hz -0.02 0.02 -0.02 0.02 f2 Hz -0.02 0.02 -0.02 0.02 Za p.u 0.69 1 0.71 1 Za ph 175.15 156.71 174.52 156.71 Zb p.u 0.74 1 0.76 1 Zb ph -190 -247 -192 -247 Zc p.u 1.16 1 1.16 1 Zc ph 183.92 157.37 182.45 157.37 Output 0 0.05 0 0.05

Table 3 Testing data of the Fault Classification Unit and their Outputs Tf 0.004 0.004 0.01 0.01 0.006 0.006 Rf 0 0 25 25 100 100 Df p.u 0.15 0.15 0.25 0.25 0.45 0.45 Fault type A-B A-B-G B-G A-B-C-G C-A-G A-B-C Io p.u 1.96 198.65 53.36 18.64 6.3 1 Za p.u 0.03 0.02 0.9 0.06 0.06 0.74 Za ph -7.1 4 134.5 37.3 69.2 34.2 Zb p.u 0.02 0.02 0.59 0.19 1 1.08 Zb ph 86.2 65 37.3 33.1 141.5 49.4 Zc p.u 0.93 0.67 0.95 0.09 0.08 1.16 Zc ph -166 -162 -202 6.9 -0.13 38.75 Output 2.3 3.08 1.2 4.1 2.51 4.02

Table 4: Testing data of the Fault Location Unit and their percentage errors. fault type DL TL DLG SLG DLG TL DL DLG TL DLG Za p.u 0.98 0.47 0.09 1 0.02 0.17 1 0.01 0.17 0.06 Zb p.u 2.59 0.67 1.02 1.02 0.02 0.21 4.95 0.01 0.21 1 Zc p.u 1.38 0.61 0.14 1.67 0.7 0.19 1.03 0.8 0.2 0.09 Df p.u 0.76 0.56 0.79 0.68 0.15 0.4 0.51 0.05 0.46 0.48 % Error 1.2 5.4 3.6 5.8 7 2.5 0.5 2.8 8 7

Tf 0.013 0.007 0.002 0.011 0.016 0.012 0.006 0.014 0.013 0.01

Rf 91 54 78 74 3 22 86 18 20 60

Za ph 142.3 21 75.4 138.3 3.2 15.4 -220 -8.7 14.3 81.3

Zb ph 147.1 25.3 138.7 142.8 81.8 -344 94.9 128.4 15.5 141.5

Zc ph -250 21.3 5.9 -228 -166 20 92.8 -141 19.8 11.9

You might also like