1 Insulation Status Assessment in High Voltage Cables Based on Decision Tree Algorithm A.R.Yazdandoust, F. Haghjoo, S.M. Shahrtash Center of Excellence for Automation and Operation of Power systems Iran University of Science & Technology, Tehran, Iran 1 Abstract--In this paper, a novel approach for insulation condition assessment is presented. This method is based on processing the detected partial discharge pulses by using a decision tree algorithm. In training procedure, not only partial discharge parameters are collected for different cavity sizes and depths, but also the influence of aging in cable insulation is considered. Results have shown high reliable performance of the proposed classifier. Also the influence of different parameters in training procedure is shown and discussed. Index Terms — Decision Tree, High Voltage Power Cables, Insulation Assessment, Partial Discharge. I. INTRODUCTION lectrical power systems include a large number of expensive and important high voltage cables of different ages, manufactured and mounted during decades. Repair and replacement of important cable system are expensive and correct scheduling for those, would give large saving of costs. Insulation system of high voltage power cable and their accessories are subjected to different kind of stress during their life, and thus to degradation and deterioration. These can lead to a reduction of life and so to a lower reliability of electrical power systems [1]. Because of the above-mentioned reasons, insulation condition assessment is very important. This paper is concerned with insulation status assessment in high voltage cables. Voids in solid insulation of high voltage XLPE cables are the most essential medium of occurring partial discharges. Growth of voids would lead to electrical trees and resulting breakdown. Different sizes and depths of these voids (i.e. the position with respect to the conductor) may cause light or severe partial discharges. Consequently, according to the partial discharge intensity, the insulation status may be categorized as Healthy, Good, Normal, Bad and A. R. Yazdandoust is with Ghods Niroo Engineering Company (G.N.E.C.), Tehran, Iran and is M.Sc. Student in Iran University of Science and Technology (IUST), Iran (e-mails: aryazdandoust@gmail.com ). F. Haghjoo is PhD student with the Department of Electrical Engineering Iran University of Science and Technology (IUST) and researcher of Relay Laboratory. (e-mails: farhadhaghjoo@gmail.com ). S. M. Shahrtash is assistant professor with the Department of Electrical Engineering, at Iran University of Science and Technology, and Director of Relay Laboratory (e-mail: shahrtash@iust.ac.ir ). Critical, i.e. from the condition that there is no partial discharge up to the severe condition that the cable needs urgent repairs or replacement. While, previously this task was performed with the assistance of fuzzy logic [2], in this paper insulation evaluation has been accomplished based on partial discharge investigation in solid insulations by using a decision tree algorithm. The aim of this paper is to classify the voids with respect to their dimensions and positions to assess the insulation status according to the remaining insulation thickness. Aging process is also considered. In order to perform this classification a decision tree with random forest algorithm is used. The input features of decision tree are apparent charge, maximum apparent charge, ratio of maximum apparent charge to apparent charge, number of discharges, quadratic rate, discharge power, and ratio of discharge power to the number of discharges; all in a predetermined number of cycles (T ref ). In training procedure, not only the above mentioned parameters are collected for different cavity sizes and depths, but also the influence of aging in cable insulation is considered. The data set is created according to the comprehensive 3-capacitors model simulation for partial discharge investigation [3], [4]. II. DECISION TREE Decision tree learning is one of the most widely used and practical methods for inductive inference. It is a method for approximating discrete-valued functions which is robust to noisy data and capable of learning disjunctive expression. Decision tree classify data sets by sorting them down the tree from the root to leaf nodes, and provides the classification of them. Each node in the tree specifies a test of some attribute of the inputs and each branch descending from that node correspond to one of the possible values of this attribute. A data set is classified by starting at the root node of the tree, testing the attribute specified by this node, then moving the tree branch corresponding to the value of the attribute. Many methods have been proposed for constructing decision tree using a collection of training and test data sets. The majority of tree construction methods use linear splits at each internal node [5]. In decision tree approach different procedures may be employed as WEKA [6], which is used in this paper. E 2008 IEEE Electrical Power & Energy Conference 978-1-4244-2895-3/08/$25.00 ©2008 IEEE