Short term load demand forecasting in Indonesia by using double seasonal recurrent Neural networks

Neural networks have apparently enjoyed con-siderable success in practice for predicting short-term hourly electricity demands in many countries. Forecasting of short-term hourly electricity in some countries usually is done by employing classical time series methods such as Winter's method and Double Seasonal ARIMA model. Recently, Feed-Forward Neural Net-works (FFNN) is also applied for electricity demand forecasting, including in Indonesia. The application of Double Seasonal ARIMA for forecasting short-term electricity load demands in most cities in Indonesia shows that the model contains both order of autoregressive and moving average. Moving average order can not be represented by FFNN. In this paper, we use an architecture of Neural Network that able to represent moving average order, i.e. Elman-Recurrent Neural Network (RNN). As a case study, we use data of hourly electricity load demand in Mengare, Gresik, Indo-nesia. The results show that the best ARIMA model for forecasting these data is ARIMA ([1,2,3,4,6,7,9,10,14,21,33],1,8)(0,1,1)²⁴(1, 1,0)¹⁶⁸. There are 14 innovational outliers detected from this ARIMA model. We use 4 different architectures of RNN particu-larly for the inputs, i.e. the input units are similar to ARIMA model predictors, similar to ARIMA predictors plus 14 dummy outliers, the 24 multiplied lagged of the data, and the combination of 1 lagged and the 24 multiplied lagged plus minus 1. The results show that the best network is the last one, i.e., Elman-RNN(22,3,1). The comparison of forecast accuracy shows that Elman-RNN yields less MAPE than ARIMA model. Thus, Elman-RNN(22,3,1) is the best method for forecasting hourly electricity load demands in Mengare, Gresik, Indonesia.