Load forecasting is necessary for economic generation of power, economic allocation between plants, maintenance scheduling and for system security. Factors that influence the peak electric load pattern are identified as seasonal variation, monthly cycles, major religious events and some unexpected random events. The goal of this paper is to model the load pattern using time series models and to compare its performance with artificial neural network model. The analysis has been structured by following a stepwise scheme, starting with the simplest model and adding each time new terms in order to assess separately the effect of the different factors that influence the daily electricity demand. The time series model will analyse the impact of exogenous variables such as maximum temperature and also seasonal dummy variables that capture the ‘day of the week’, ‘holiday’ and ‘month of the year’ effects. From the results, the influence of maximum temperature and seasonality is proved and is significant. In ANN we will used backpropagatian algorithm. Backpropagation model is basically a gradient descent method and its objective is to minimize the mean squared error (MSE) between the target values and the network outputs in this study, we consider various activation functions in the derivation of the proposed method for the backpropagation model, with proposed error function of . Training and testing of these different activation functions were carried out using electricity load data to forecast the daily maximum electricity load. Convergence rate and forecast accuracy of various activations function were then compared. Finally the one month ahead forecast of maximum electricity demand from time series models and neural network models were then compare.
Keywords: Electricity Load Demand, Time Series, Artificial Neural Network and Forecasting