IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 2, APRIL 2000
833
Abstract—An integrated fuzzy expert system is presented to diagnose various faults that may occur in a regional transmission network and substations. Fuzzy reasoning method is applied, and it is discussed in detail. Discrimination of false operations or nonoperations of protective devices as well as the fault identification scheme are also analyzed, together with the fuzzy inference process. The proposed system is designed to improve efficiency, generality, and reliability of the solution. The system will replace a fault diagnosis system that had been tested as a part of an intelligent support system on a local control center in Korea.
Index Terms—Fuzzy Reasoning, Integrated Fault Diagnosis, Expert System.
I. I NTRODUCTION
T HE PAST two decades have revealed great advances in the application of artificial intelligence technologies to power systems. Fault diagnosis of power systems is a major subject of expert systems applications. Fukui [1] and other papers [2],
[3] have proposed expert systems to diagnose faults in transmission networks. As substation automation has become an important topic, fault diagnosis systems for substations [4]–[9] have been proposed. Protopapas [4] proposed an interactive expert system for nonexperts using PC-SCADA, Venkata [5] proposed a basic expert system approach on the ICPS (Integrated Control and Protection System), and Japanese power corporations have announced prototype or practical fault diagnosis systems
[6]–[9] for substations. Kansai Electric Power Corporation has developed one for a 500 kV substation [6], and another one for a 275 kV substation [7]. Tokyo Electric Power Corporation has developed a supervisory system [8], and Chubu Electric Power
Corporation has presented a frame-based expert system [9].
Regarding the application of uncertain reasoning methods to the fault diagnosis, The Certainty Factor approach [10] is the first application for the diagnosis of transmission systems.
Fuzzy approaches for substation diagnosis [11] and for the diagnosis of transmission Networks [12] were proposed. However the structure of the latter system is too complex to be applied to
Manuscript received June 1, 1998; revised June 30, 1999. The authors wish to acknowledge the financial support of Korea Research Foundation made in the program year of 1997.
H. J. Lee is with Kwangwoon University, Korea (e-mail: hjlee@daisy.kwangwoon.ac.kr).
D. Y. Park and B. S. Ahn are with the Research and Development Center, LG
Industrial Systems Co., Korea.
Y. M. Park and J. K. Park are with the School of Electrical Engineering, Seoul
National University, Korea.
S. S. Venkata is with Iowa State University, USA (e-mail: Venkata@iastate.edu).
Publisher Item Identifier S 0885-8977(00)03513-5.
practical systems. This paper introduces an integrated fault diagnosis fuzzy expert system to diagnose various faults including multiple faults that may occur in a transmission network and substations. Discrimination of false operations and nonoperations of protective devices, and the fault identification method are also analyzed, together with fuzzy reasoning.
II. P ERFORMANCE OF D IAGNOSIS S YSTEMS
Regarding the fault diagnosis problem, in general, the performance of expert systems can be evaluated by focusing on three major aspects—reliability, generality, and efficiency.
As pointed out in references [10], [11], since the diagnostic process is an inductive reasoning process in nature, it is desirable to apply an inexact reasoning process, although it is also possible by deterministic reasoning processes using heuristic knowledge. Regardless of reasoning methods, the reliability should be checked by induction. Actually, this is a major reason why deductive or numerical algorithmic approaches are hard to apply to this area. The reliability of solutions in the proposed system will be discussed in Section III.
A second point is the generality, that means herein the wide applicability. It is important and desirable that a fault diagnosis system for a substation can be easily applied to many other substations that have the same structure without modifying the rule-base. Because it is clear that human diagnostic knowledge is only dependent on the structure itself, such as “double bus–double breaker structure” or “breaker and a half structure,” regardless of the number of transformers or distribution feeders.
A topology based knowledge representation method [13], [14] is applied to this system to achieve this goal. For example, a transmission line in a substation is represented as in reference
[14].
capacity; t-type
(1) where ss name t1 name denotes a name of a transmission line rrr xxx ccc capacity is a resistance value of the transmission line is a reactance value of the transmission line is a capacitance value of the transmission line is a capacity of the transmission line in [MVA] denotes the type of the transmission line such as
Ground or Overhead
The proposed system has been developed for substations that have the “double bus–double breakers structure,” a typical structure in Korean electric power systems. However it
0885–8977/00$10.00 © 2000 IEEE

834 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 2, APRIL 2000 will be extended to cope with other structures for worldwide application.
The third criterion is the efficiency or the processing time.
This is fairly dependent on the applied reasoning method, searching strategy, and data structure. It takes a maximum of one second for this system to generate feasible solutions even in the most complex cases. This behavior results from the modular structure and application of a meta-inference scheme.
The structure will be introduced later.
So far the system has been tested in a local control center that controls 9 substations and 23 transmission lines. It has demonstrated that the processing time is not dependent on the size of the network and the number of substations in the system. This feature results from the topology-based rule-base as discussed in [11], [14].
Fig. 1.
Sample power system.
III. F UZZY R EASONING
A. Uncertainty in the Power System Fault Diagnosis
When a fault occurs in a power system, information of operated protective devices is transferred to the SCADA system.
It is known that there are causal relationships between a fault and the corresponding protective relay or relays, and between a relay and the corresponding circuit breakers [11]. The causal relationship is represented as “if-then” rules as follows. Rule 1 represents a correct operation of a relay or relays, and rule 2 represents a nonoperation.
Rule 1: If a fault occurs, then the corresponding relay or relays may operate.
Rule 2: If a fault occurs, then the corresponding relay or relays may not operate.
Since these two rules represent deductive uncertain knowledge, if the probability of a nonoperation is known, then it is possible to forecast the future behavior of protective devices for a given fault. Therefore, this knowledge can be applied to a fault simulator.
Meanwhile, from a diagnostic viewpoint, only the conclusion parts of rules are known using alarm information. Therefore it is basically impossible to identify the real fault sections by deductive reasoning, which means that the fault diagnosis problem is an inductive reasoning process in nature. Because a premise should be inferred using the information of a conclusion in a rule, there exist uncertainties in the fault diagnosis and this is the main reason why the fuzzy reasoning method is applied to this system.
Operations of protective devices can be classified into four categories—correct operation, false operation, correct nonoperation, and nonoperation. Again, as the diagnosis is an inductive reasoning process, the assumption of one or more fault sections is inevitable. This means that the discrimination of a false or the nonoperation of a relay can be determined under the assumption of a fault section or sections. Furthermore, the possibility of a relay operation is supported by the operation of circuit breakers tripped by the relay, and the possibility of nonoperation of a circuit breaker is supported by operation of the corresponding backup protective relay.
As an example, Fig. 1 shows a simple network that consists of three substations and two transmission lines.
A fault occurred in a substation. Two relays and three circuit breakers operated as follows.
87B I at Alpha substation (bus differential relay)
Z2 at Gamma substation (back-up distance relay)
CB1 and CB4 tripped at Alpha substation
CB7 tripped at Gamma substation
Among the solutions listed in Table I, the two most possible hypotheses are shown below without loss of generality. Actually the system generates six solutions in order of possibility among the nine solutions in Table I, because the system is currently set to find solutions whose possibilities are greater than
0.6. In fact, within our test, a solution whose possibility is below
0.8 is meaningless. So far, as the real fault section or sections have been exactly estimated by the first solution for all cases, the other solutions have no meaning. Herein it is noted that a possibility, a kind of “subjective probability,” is a measure of uncertainty in fuzzy theory.
1) Possible solution 1 (Possibility = 0.95): A fault occurred on the BUS 1 at Alpha substation.
87B1 operated.
CB1 and CB4 tripped.
CB3 did not operate.
Possible solution 2 (Possibility = 0.69): Double faults occurred on Bus1 and Bus2 at Alpha substation.
87B1 operated
CB1 and CB4 tripped.
CB3 did not operate.
Z2 at Gamma substation operated
CB7 tripped.
87B2 did not operate.
Herein, it should be noted that there is no uncertainty in the operations themselves of the protective devices. Actually the uncertainty exists in the reasoning process of what hypothesis is the most possible. It is clear, in this example, that the nonoperation of “87B2” can be evaluated as the “correct nonoperation” in the first solution, but it is evaluated as the “nonoperation” in the second solution. The final step is to choose the most possible hypothesis among candidates.

LEE et al.: A FUZZY EXPERT SYSTEM FOR THE INTEGRATED FAULT DIAGNOSIS
TABLE I
D IAGNOSIS R ESULTS FOR THE E XAMPLE IN F IG . 1
835
In the concept of possibility [17], a rule representing an operation of protective device can be translated into a fuzzy proposition using the possibility as the possibility of a fuzzy proposition, can be thought of a fuzzy truth-value of the rule. Therefore if possibility values are assigned to propositions, then several rules are combined to extract the conclusive possibility of the hypothetical fault section using fuzzy reasoning.
In the next subsection, the causal relationships among the operations of protective devices will be defined as fuzzy relations, and the membership values of the fuzzy relations will be evaluated. Finally, it will be shown that the membership value can be used directly for the value of possibility.
TABLE II
M EMBERSHIP V ALUES OF F UZZY R ELATIONS OF P ROTECTIVE D EVICES (a)
M EMBERSHIP V ALUES OF F UZZY R ELATIONS B ETWEEN THE O PERATION OF A
R ELAY AND A F AULT , (b) M EMBERSHIP V ALUES OF F UZZY R ELATIONS
B ETWEEN THE O PERATION OF A C IRCUIT B REAKER AND THE O PERATION OF A
R ELAY , (c) M EMBERSHIP V ALUES OF F UZZY R ELATIONS B ETWEEN THE
O PERATION OF A B ACKUP R ELAY AND THE N ONOPERATION OF D EVICES
B. Fuzzy Relations of the Protective Device Operations
Individual membership value for each fuzzy relation, as discussed in the previous subsection, may be chosen by the analysis of past fault history in power systems. Heuristics of experienced operators can also be reflected to some extent. Five membership values of fuzzy relations are assigned, as shown in Table II, to represent the causal relationships. Besides, as the possibility of multiple faults is lower than that of single fault, another value
1 is assigned, and it is illustrated in Fig. 2.”
The equation (2) is claimed here, for example, in case of a fuzzy relation between the correct operation of a relay and a fault.
where, (Correct Operation of Relay, Fault) is a membership value of the fuzzy relation, a relay and a fault defined in Table I, and
P os(P ) of the fuzzy proposition,
PP
R
P os(P ) is a possibility
P , such that “The cause of correct operation of a relay may be a fault.
The diagnostic process is described by chained fuzzy relations as shown in Fig. 2. In this figure, each numeral on an arc denotes a membership value of each ftizzy relation. A shaded area denotes alarms that have been received from the remote terminal unit of a substation.
C. Fuzzy Reasoning Processes of Fauft Diagnosis process
In the previous subsection, it was shown that rules representing operations of protective devices can be defined as fuzzy relations and that the possibility value of a fuzzy proposition about the cause of an operation of protective device equals the membership value of the fuzzy relation. Consequently, the possibility value of a hypothetical fault section can be evaluated using the fuzzy reasoning process. As a method for composition of rules in fuzzy reasoning, the max-product composition method [18] is applied here for simple operation.

836 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 2, APRIL 2000
Fig. 2.
Fuzzy representation of fault diagnosis process.
1) Max-Product Composition of Fuzzy Relations: If two binary fuzzy relations,
R by their respective membership values as in (3) and (4) then the composition of these relations,
Q = R by their membership values as in (5)
R (3) where
R
(xxx; yyy) (X1Y) is a membership value of fuzzy relation where,
SSS(yyy; zzz) yyy 2 YYY ; zzz 2 ZZZ
(yyy; zzz) is a membership value of fuzzy relation SSS
(4)
(xxx; zzz) = Max[ (xxx; yyy) 2 (yyy; zzz)]
(5)
Fig. 3.
Overall structure of the expert system.
where,
Max means a logical OR operation, and ’×’ means a logical AND operation.
As illustrated in Fig. 2, the reasoning process that calculates the possibility of a fault section is as follows.
Step 1: Calculate the fuzzy membership value for the compositional fuzzy rule in the path determined by matching the information of operated protective devices. In this step, the protective devices in the path that did not operate are identified as the nonoperated devices.
Step 2: Calculate the fuzzy membership value for the path of a falsely operated relay that has nothing to do with the fault.
Step 3: Calculate the overall membership value for the hypothetical fault section by joining step 1 and step 2. This membership value is used for the possibility value of the fault section.
In this way, all the hypothetical fault sections are assigned their own possibilities and the fault section can be estimated by choosing the hypothesis with the greatest possibility. The nonoperation and the false operation of each protective device are easily identified in this process. Furthermore it is noted here that the sequence of operation is also identified, therefore sequence information is used for the explanation of the solution. It is an important point that, as shown in the example in Fig. 1, the sequence can be inferred although the alarms have no time information at all. SOE (Sequence Of Event) information may improve the performance of the system. However this information is currently not used in the control center where the system is installed.
IV. S TRUCTURE OF THE E XPERT S YSTEM
The proposed fault diagnosis expert system consists of four major expert systems [11], [14], [16] as shown in Fig. 3.

LEE et al.: A FUZZY EXPERT SYSTEM FOR THE INTEGRATED FAULT DIAGNOSIS 837
A. Meta Inference System
The meta-inference system coordinates the operation of the other three expert subsystems, and controls the order of program execution using meta-knowledge [19]. Actually it analyzes the alarms first, then decides an expert system to be run when a fault occurs. In normal states, it turns over control to its own monitoring program [16].
one second, even in the most complicated cases. So far many case studies, including the past history of faults in the domestic system have demonstrated the 100% diagnostic performance of the system, which means that the real fault sections have been exactly estimated as most possible solutions generated by the system for all cases. Since multiple faults have not occurred yet in the KEPCO system, human experts in KEPCO provided the data to test multiple fault cases.
B. Expert System for the Hybrid Diagnosis
The hybrid diagnosis expert system is activated when faults occur in a substation, and the blackout region is extended to the outside of the substation by the nonoperation of corresponding protective relays or circuit breakers. Faults at high voltage buses or transformers may cause the situation. In these cases, the system is activated by the meta-inference system, and it uses a partial database and rule base of the other two expert systems. The above example in Fig. 1 is this case, since the blackout regions are the Alpha substation and transmission line
2. The system uses a database for Alpha substation and data of line 2 from the network database. Then it applies the hybrid rule base.
C. Expert System for the Diagnosis of Substations
The system diagnoses faults that occurred inside one of the substations, and when they were cleared inside the substation
[11] guided by the meta-inference system. As shown in Fig. 3, the system has a common rule-base since the rules are generally represented based on the topology. There are nine databases for nine substations now. As mentioned before, the number of transformers or distribution lines are different in each substation.
Seven protective relays are considered as follows:
1) Over-current relays
2) Bus protective relay
3) Transformer differential relay
4) Lockout relay
5) Buchholtz relay
6) Pressure valve
7) Sudden pressure relay
D. Expert System for the Diagnosis of Transmission Network
The system diagnoses faults that occurred in the transmission network. Herein, it is noted that a fault that occurred in a transmission line can be cleared by the line protective relays. Typical zone relays are considered since Z1 relays are installed as the primary protection, and Z2, Z3 relays as the backup protection in the secondary 154kV transmission network in KEPCO system.
V. C ONCLUSIONS
An on-line fuzzy fault diagnosis expert system has been developed to assist SCADA operators in local control centers. The system diagnoses various faults occurring in several substations and the transmission network using the fuzzy reasoning process.
Discrimination of false operations and nonoperations of protective devices was discussed in detail, together with the fuzzy reasoning process. The system can diagnose all faults within
A CKNOWLEDGMENT
The authors wish to express special gratitude to the researchers of Korea Electric Power Research Institute for their support of practical data and comments.
R EFERENCES
[1] Chihiro Fukui and Junzo Kawakami, “An Expert System for Fault Section Estimation using Information from Protective Relays and Circuit
Breakers,” IEEE Trans. on Power Delivery, vol. PWRD-I, no. 4, pp.
83–90, Oct. 1986.
[2] Kevin Tomsovic, Chen Ching Liu, Paul Ackerman, and Steve Pope, “An
Expert System as a Dispatchers’ Aid for the Isolation of Line Section
Faults,” IEEE Trans. on Power Delivery, vol. 2, no. 3, pp. 736–743, July
1987.
[3] Takafumi Kimura, Sinya Nishmatsu, Yoshiteru Ueki, and Yoshikazu
Fukuyama, “Development of an Expert System for Estimating Fault
Section in Control Center based on Protective System Simulation,”
IEEE Trans. on Power Delivery, vol. 7, no. 1, pp. 167–172, Jan. 1992.
[4] C. A. Protopapas, K. P. Psaltiras, and A. V. Machias, “An Expert System for Substation Fault Diagnosis and Alarm Processing,” IEEE Trans. on
Power Delivery, vol. 6, no. 2, pp. 648–655, April 1991.
[5] B. Jeyasurya, S. S. Venkata, S. V. Vadari, and J. Postforoosh, “Fault Diagnosis using Substation Computer,” in Proceeding of CIGRE’89, pp.
289–295.
[6] Shunichi Ito, Isao Hata, Taizo Hasegawa, Masahiko Amano, and Akira
Maruyama, “Advanced Operation Guidance Expert System for 500kV
Substation,” in Third Symposium on ESAP, Japan, 1991, pp. 405–412.
[7] Kazuo Harnamoto, Masakazu Osada, Haruki Oue, Kuniomi Yamada,
Yoshiyuki Takaoka, and Fumio Higashiyama, “275KV Substation Operation Support System,” in Third Symposium on ESAP, Japan, 1991, pp. 419–426.
[8] Syoichi Muto, Tetsuo Matsuda, Yoshihumi Yamagata, Yasuo
Kawakami, and Yuji Tanaka, “Supervisory System for Substation with Expert System,” in Third Symposium on ESAP, Japan, 1991, pp.
413–418.
[9] S. Kumano, H. Ito, T. Goda, U. Uekubo, S. Kymoto, H. Kourogi, and Y.
Ariura, “Development of Expert System for Operation at Substation,”
IEEE Trans. on Power Delivery, vol. 8, no. 1, pp. 56–65, Jan. 1993.
[10] Y. M. Park and H. J. Lee, “An Expert System for the Fault Diagnosis in
Power Systems,” in Proc. Of IFAC on Power System and Power Plant
Control, Korea, Aug. 1989, pp. 697–701.
[11] H. J. Lee, B. S. Ahn, and Y. M. Park, “A Fault Diagnosis System for Distribution Substations,” IEEE Trans. on Power Delivery, to be published.
[12] H. J. Cho, J. K. Park, and H. J. Lee, “A Fuzzy Expert System for Fault
Diagnosis of Power Systems,” in Proc. of ISAP’94, France, Sep. 1994, pp. 214–222.
[13] H. J. Lee, Y. M. Park, and J. K. Park, “Investigation of Effective
Topology Representation for Substations,” Trans. of KIEE, vol. 45, no.
7, pp. 921–924, Jul. 1996.
[14] Heung-Jae Lee and Young-Moon Park, “A Restoration Expert System for 154kV Distribution Substations,” Trans. on Power Delivery, vol. 11, no. 4, pp. 1765–1770, Oct. 1996.
[15] K. P. Adlassing, “Fuzzy Set Theory in Medical Diagnosis,” IEEE Trans.
on System Man and Cybernetics, vol. 16, pp. 260–265, 1986.
[16] H. J. Lee, Y. M. Park, K. H. Lee, Y. B. Yoon, and J. B. Choo, “Development of an intelligent support system at local control centers in
KOREA,” in Proc. of 1CPST94, Beijing, China, 1994, pp. 1056–1060.
[17] G. J. Klir and B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Appli-
cations: Prentice Hall International, Inc., 1995.
[18] E. H. Mamdani and B. R. Gaines, Eds., Fuzzy Reasoning and its Appli-
cations: Academic Press, 1981.

838 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 15, NO. 2, APRIL 2000
[19] D. A. Waterman, A Guide to Expert Systems: Addison-Wesley Publishing Co., 1986.
Heung-Jae Lee received B.S., M.S., and Ph.D. degrees from Seoul National
University, in 1983, 1986 and 1990, respectively, all in electrical engineering.
His major research interests are expert systems, neural networks, and fuzzy systems applications in power system operation areas including computer applications. He is an associate professor at Kwangwoon University.
Young-Moon Park received the B.S., M.S., and Ph. D. degrees from Seoul
National University, in 1956, 1959 and 1971, respectively, all in electrical engineering. His research interests are power systems control and planning. From
1989 to 1991, he was the director of KIEE, and from 1992 to 1996, he was the director of EESRI. He is an IEEE fellow and an emeritus professor at Seoul National University.
Deung-Yong Park received B.S. and M.S. degrees from Seoul National University, in 1992 and 1994, respectively, both in electrical engineering. His major research interests are expert systems, neural networks and fuzzy systems applications to power systems. He is a research engineer of Power Systems Laboratory in the Research and Development Center of LG Industrial Systems Co.,
Ltd.
Jong-Keun Park received a B.S. degree from Seoul National University in
1973, and the M.S. and Doctor of Engineering degree from Tokyo University in
1979, 1982, respectively, all in electrical engineering. His research interests are power systems control and planning, and AI applications to power systems. He is an IEEE Senior Member and a full professor at Seoul National University.
Bok-Shin Ahn received the B.S., M.S. and Ph.D. degrees from Seoul National
University, in 1980, 1982 and 1998, respectively, all in electrical engineering.
His major research interests are the computer application to power systems and distribution automation. He is a head of Power Systems Laboratory in the Research and Development Center of LG Industrial Systems Co., Ltd.
S. S. Venkata received a B.S. degree from Andhra University, India in 1963, an
M.S. degree from the Indian Institute of Technology, Madras, India in 1965, and a Ph.D. degree from the University of South Carolina in 1971, all in electrical engineering. He is a fellow of IEEE, and a full professor and Chairman of the
Department of Electrical and Computer Engineering at Iowa State University.