Residential Lighting Load Profile: ANFIS and Neural Network-Based Models

Abstract

This study presents methodologies (ANFIS and Neural Network-based models) based on characterization of variables that impact on lighting usage and has the platform of addressing and solving non-linear issues, ambiguity and randomness of data associated with lighting usage and models the lighting demand according to time of use (TOU) periods. Variables considered in the development of the models include natural lighting, occupancy (active) and income level. During the training process of the ANFIS-based and NN-based model trapezoidal membership and sigmoid transfer function were applied respectively. The ANFIS-based model interpreted the complexity associated with lighting usage, learned and adapted historical patterns and computed its output based on the associated characterizations than NN-based method. The ANFIS -- based model showed good prediction accuracy in the time of use period (TOU) analysis especially standard and peak periods for lighting demand. This is very important for electricity distribution planners, energy conservation project evaluation and implementation etc.