Descriptive statistics is the traditional method used to analyse the meter reading information from domestic’s smart metering system become difficult because of increasing available amount of data of electricity utilities due to implementation of smart metering system. Particularly when aggregation and average process are applied, key characteristic information’s are lost often. To overcome this problem some other data analysing method is needed so that loss of information is eliminated. Data mining is the one method which provides solution by analysing large amount of information’s and also this method has proper approach, before aggregation processes are applied data to be segmented. Dimension reduction achieved by segmentation process and also it leads to easy manipulation of data. In an electricity industry clustering method has been used for an indefinite period and however the application of clustering method in domestic level restricted to date. This research work describes three most widely used clustering methods such as self-organizing maps (SOM), K-Means, k-medoide. Pattern of household items is determined against electricity usage across the day for an individual household, then segmentation process to be carried to cluster the data according to the pattern from that Best performing technique is evaluated. The procedure is rehashed for every day over a six month time frame in order to identify the intra daily, diurnal, seasonal variation of electricity demand. Series of profile classes generated based on the results and the patterns shows common pattern of the electricity consumed with in the home 2.
First step is to find the suitable clustering methodology for the segmentation process by evaluating available methods. Three commonly used algorithm analysis methods (Self organizing maps,K means medoide) were investigated for electricity industry then for the segments, number of clusters were investigated. Appropriate number of clusters and clustering methods identified with the help of Davies Bouldin validity index method. This is the common method to evaluate the quality of the data sheet and the index was evaluated over the period of three random days. Final stage of the process is to cluster each day separately on 24 hours basis over the time of six months if the number of clusters and clustering method is identified.
Electricity load profile of cluster per day is determined by calculating electricity demand for individual cluster on a particular day by average method. Clusters slightly differed in terms of both timing of electricity used and magnitude were combined together to form series of profile classes and the size were small. Each class representing electricity used half hourly of each day over the period of six months respectively and the vector size is 48*184 data points. Customer class index recorded each customer electricity usage on a particular day. Which PC used majority of the time over the period of six months is determined with the help of statistical mode of customer class index 2.