Energy forecasting plays a dominant role in the sustainable development economic optimization,
resource planning and secure operation of electric power systems. The variation in energy demand is a
major source of uncertainty in planning for future capacity enhancement, resource needs and operation of
existing generation resources. Electric utilities need monthly peak and yearly demand forecasting for
budgetary planning, maintenance scheduling and fuel management. The software tools currently used by
the utilities and the various regulatory organizations indicate that ‘official long term energy demand’
forecasts of these systems are based, at best, on some form of linear or log-log linear regression
(econometric) models with the parameters often estimated using the ordinary least-squares method. These
approaches aim to develop mathematical models on the basis of available data. Most of the utilities have
attempted to build energy forecasting models based on different behavioural assumptions about the shape
of the demand curve. The parameters of these models are estimated by a variety of techniques. However,
the in-depth study of energy demand data carried out by the authors has revealed its chaotic nature. Hence,
the conventional modeling techniques have resulted in limited success in the forecasting and modeling of
energy demand. This paper aims to depict the chaotic nature of energy demand by carrying out
experimental studies on U S energy data and comparing its distinguishing features such as Correlation
Dimension, Embedding Dimension and False Nearest Neighbours, Lyapunov Exponents and Predictive
Power, Return Map and Power Spectral Density with those of a benchmark chaotic systems, viz., Box and
Jenkins gas furnace data. The results clearly indicate that energy demand should be treated as chaotic
system for better efficacy in its modeling and prediction.