672 J SCI IND RES VOL 69 SEPTEMBER 2010 Journal of Scientific & Industrial Research Vol. 69, September 2010, pp. 672-679 *Author for correspondence E-mail: aazadeh@ut.ac.ir; ali@azadeh.com An integrated ANN-K-Means algorithm for improved performance assessment of electricity distribution units Ali Azadeh 1* , Morteza Saberi 2,4 Mona Anvari 3 and M Moghaddam 1 1 Department of Industrial Engineering, Department of Engineering Optimization Research, Research Institute of Energy Management and Planning, and Center of Excellence for Intelligent Experimental Mechanics, Faculty of Engineering, University of Tehran, PO Box 11365-4563, Iran 2 Department of Industrial Engineering, University of Tafresh, Tafresh, Iran 3 Department of Industrial Engineering, Iran University of Science and Technology, Iran 4 Institute for Digital Ecosystems & Business Intelligence, Curtin University of Technology, Perth, Australia Received 05 April 2010; revised 24 June 2010; accepted 28 June 2010 This study proposes a non-parametric efficiency frontier analysis method based on artificial neural network (ANN) and K-Means algorithm for measuring efficiency of electricity distribution units (EDUs). Effect of return to scale of EDU on its efficiency is included and EDU used for correction is selected based on its scale. K-Means algorithm is used to cluster EDUs to increase their homogeneousness by handling outlines and noise. Proposed approach was applied to 31 EDUs in Iran . This is first study using integrated ANN-K-Means algorithm for improved performance assessment of EDUs. ANN-K-Means algorithm is compared with earlier models to show its advantages and superiorities in prediction and forecasting. Keywords: Artificial neural network, Electricity distribution units, Improved performance assessment, K-Means algorithm Introduction Efficiency frontier analysis (EFA) is an important approach of evaluating firms’ performance. A significant number of methodologies with different approaches (parametric and non-parametric) and methods are reported 1 to characterize such efficiency. Parametric approaches (PAs) include estimation of both deterministic and stochastic frontier function (SFF), which are based on econometric regression theory. Non- parametric approaches (NPAs) include Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH), which are based on a mathematical programming approach. In all methodologies, frontier is defined by most efficient decision making unit (DMU) of sample. Mathematically, frontier methods are introduced as a high-reliability analysis tool and have been largely used for studies in electrical and industrial field 1,3 . A number of EFA methods 1,4-14 make assumption about functional form of frontier, whereas NPAs do not make such assumption. In electricity restructuring and reform, performance assessment and estimation efficiency scores of electricity distribution is a very important issue for regulators. Efficiency scores are used to set X-factors in incentive regulation methods 12-15 (price cap and revenue cap regulation). In regulation and evaluation of electricity distribution units (EDUs), DEA, Corrected Ordinary Least Squares (COLS) and Stochastic Frontier Analysis (SFA) are applied. DEA uses linear programming to calculate efficiency in a given set of DMUs. COLS is an alternative frontier-oriented approach for measuring relative efficiency of units to estimate best practice frontier and efficiency scores. Among several studies on efficiency estimation and benchmarking of EDUs, Goto & Tsutsui 16 measured overall cost efficiency and technical efficiency between Japanese and US electricity utilities by DEA. Försund & Kittelsen 17 applied DEA efficiency scores to measure Malmquist productivity index in Norwegian electricity distribution companies 17 . Resende 18 used non-parametric input-oriented DEA model for evaluation of Brazilian EDUs. Edvardsen & Försund 19 studied performance of 122 electricity distributors in Denmark, Finland, Norway, Sweden and Netherlands for 1997. Giannakis et al 20 found that cost-efficient firms do not necessarily exhibit high service quality, and efficiency scores of