Dynamic Stochastic Model to Forecast Non Stationary Electricity Demand
Mohammad Anwar Rahman
Mohammad Anwar Rahman, Assistant Professor at the University of Southern Mississippi (USM), USA.
Manuscript received on July 24, 2013. | Revised Manuscript received on August 10, 2013. | Manuscript published on August 30, 2013. | PP: 272-277 | Volume-2, Issue-6, August 2013. | Retrieval Number: F2073082613/2013©BEIESP
Open Access | Ethics and Policies | Cite
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: This paper presents a dynamic stochastic model to forecast the pattern of residential electricity consumption of a rapidly developing industrial nation. Electricity usage is essential for continuous economic development and urbanization. Long term projection of residential electricity demand is vital for decision makers to develop strategic resource planning and energy policy. In this forecasting model, electricity demand is a function of the price of electricity, household electric appliances, real personal income, number of households, and urban conditions. We propose the Bayesian statistical technique on a dynamic linear model to predict the parameters of the demand model. We apply the model to a time series of a nonlinear, non-stationary household electricity demand. The forecast is generated from the inference of marginal posterior distribution of the model parameters obtained with a Markov Chain Monte Carlo simulation method. The forecast result is tested and compared with actual data and two alternate models. The Bayesian model is proven to be an effective forecasting method with the flexibility to solve multi-dimensional time series models and update estimated parameters as the demand changes over time. Test results indicate that Bayesian model is preferred over the classical artificial neural networks and the regression models due to its capacity to predict parameters of large-scale multivariate models.
Keywords: Bayesian statistical model, Classical artificial neural network, Dynamic linear model, Electricity load data, forecast validation.