Predicting Electricity Usage of Houses Incorporating Smart Meters

Abstract

The need for smart grid user-side  management is becoming more critical as smart power  distribution networks evolve. This study suggests a short term power load forecasting model based on K-means and  Clustering to increase the precision of short-term electric  load forecasting for individual customers. Local  comparable daily data are used as characteristics for users  with poor correlation at adjacent times. Based on the - means cluster analysis technique, a clustering module was  created. The clustering module was validated using home  load profiles based on smart meters. Utilizing daily and  segmented load profiles from individual and collective  smart meter data, many case studies were put into practice.  The smallest segmentation time window yields the lowest  clustering ratio, according to simulation findings described  in terms of the correlation between the clustering ratio and  the time window. Results also indicate that for strongly  coupled load profiles, a limited number of clusters is  advised. Last but not least, the feather vectors are sent into  the BP Neural Network, which forecasts the short-term  load. According to experimental findings, the suggested  clustering algorithm matches the features of customers'  power consumption pattern. The accuracy of clustering based load forecasting is higher than that of non-clustering  load forecasting when using the same forecasting  methodology. This model offers superior predicting  accuracy when compared to conventional BP, RBF, and  GRNN Neural Network