1284 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 26, NO.
2, APRIL 2011
Improving SVM-Based Nontechnical Loss Detection in Power Utility Using
the Fuzzy Inference System Jawad Nagi, Keem Siah Yap, Sieh Kiong Tiong, Member, IEEE, Syed Khaleel Ahmed, Member, IEEE, and Farrukh Nagi
Abstract—This letter extends previous research work in mod-
eling a nontechnical loss (NTL) framework for the detection of fraud and electricity theft in power distribution utilities. Pre- vious work was carried out by using a support vector machine (SVM)-based NTL detection framework resulting in a detection hitrate of 60%. This letter presents the inclusion of human knowl- edge and expertise into the SVM-based fraud detection model (FDM) with the introduction of a fuzzy inference system (FIS), in the form of fuzzy IF-THEN rules. The FIS acts as a postprocessing scheme for short-listing customer suspects with higher probabili- ties of fraud activities. With the implementation of this improved SVM-FIS computational intelligence FDM, Tenaga Nasional Berhad Distribution’s detection hitrate has increased from 60% to 72%, thus proving to be cost effective. Index Terms—Computational intelligence system, fuzzy logic, nontechnical loss, pattern classification.
I. INTRODUCTION
ISTRIBUTON losses in power utilities originating from
PARAMETERS SELECTED FOR FORMATION OF THE CUSTOMER SELECTION RULE COMPARISON OF THE DETECTION HITRATE FOR SVM AND SVM-FIS FRAMEWORK
E. Evaluation of the FIS
The FIS [4] is implemented by evaluating the fuzzy rule in Table II for all customers (FIS testing samples) shortlisted as fraud by the SVM in [2]. The FIS produces a fuzzified output TABLE II value in between 0 to 1 for each testing sample (customer). FORMATION OF ORDINARY AND FUZZY RULE FOR CUSTOMER SELECTION Testing samples with a fuzzy output value of greater than 0.5 (default threshold in between 0 to 1) are considered as customers with higher probabilities of fraud activities and abnormalities.
III. EXPERIMENTAL RESULTS
Pilot testing results obtained from TNBD Sdn. Bhd. for onsite customer inspection by using the SVM-based FDM in [2] and the SVM-FIS FDM are indicated in Table III. The obtained results indicate that the computational intelligence scheme of SVM-FIS outperforms the SVM-based FDM in [2] by contributing a 12% increase in the average detection hitrate.
high probabilities of fraud and abnormalities, as indicated in IV. CONCLUSION
Table I. The approach proposed in this letter extends the research work conducted in [2] by integrating human knowledge and B. Formation of the Customer Selection Rule expertise into the SVM-based FDM with the implementation The parameters selected in Table I are analyzed in order to de- of an FIS. The FIS first introduced in [4] can be used as a velop an ordinary rule for the selection of customers with high decision-making system for various applications. To the best probabilities of fraud activities and abnormalities. The customer of our knowledge, the work presented here is the first to use selection rule is established by using discriminative features in the FIS for detection of fraud and electricity theft in power order to classify (identify) consumption patterns of good cus- distribution utilities. tomers [2, Fig. 4] against consumption patterns fraud customers In this letter, the FIS acts as a postprocessing scheme to short- [2, Fig. 3], as shown in Table II. The customer selection rule fil- list customer suspects with high probabilities of fraud and ab- ters (removes) customers with low probabilities of fraud activi- normalities. The FIS emulates the reasoning process that human ties from the shortlisted suspects. experts (TNBD NTL Engineers) undertake in detecting fraud activities. The higher detection hitrate of the SVM-FIS proves C. Transformation of the Ordinary Rule Into Fuzzy Rule it to be more cost effective than the FDM in [2]. With the im- plementation of the SVM-FIS-based FDM, TNBD’s detection In order to implement fuzzy reasoning, the ordinary rule in hitrate will increase 12% from 60% to 72%. Table II is transformed into a fuzzy IF-THEN rule, using defini- tions for combining fuzzy sets. For union operations (OR), the REFERENCES fuzzy set “ ” operation is used and for intersection opera- [1] A. H. Nizar and Z. Y. Dong, “Identification and detection of elec- tions (AND), the “ ” operation is used. tricity customer behaviour irregularities,” in Proc. IEEE/Power Eng. Soc. Power Systems Conf. Expo., Seattle, WA, Mar. 15–18, 2009, pp. D. Fuzzy Membership Function Formulation and Tuning 1–10. [2] J. Nagi, K. S. Yap, S. K. Tiong, S. K. Ahmed, and M. Mohamad, “Non- Triangular and trapezoid membership functions (MFs) using technical loss detection for metered customers in power utility using the ten parameter values defined in the ordinary rule in Table II support vector machines,” IEEE Trans. Power Del., vol. 25, no. 2, pp. 1162–1171, Apr. 2010. are used to implement the fuzzy rule in Table II. The values of [3] A. H. Nizar, Z. Y. Dong, and Y. Wang, “Power utility nontechnical loss the MFs are determined by inspecting load profiles of customers analysis with extreme learning machine model,” IEEE Trans. Power confirmed as fraud by TNBD onsite inspection teams. The MFs Syst., vol. 23, no. 3, pp. 946–955, Aug. 2008. [4] J.-S. R. Jang, “ANFIS: Adaptive-network-based fuzzy inference are fine tuned by using knowledge and expertise from engineers system,” IEEE Trans. Syst., Man, Cybern., vol. 23, no. 3, pp. 665–685, in TNBD’s NTL Group. May/Jun. 1993.