Weekly Load Prediction Using Wavelet Neural Network Approach

Electric load prediction has drawn attention of many researchers due to its prerequisite concern for accurate scheduling, planning and operations of electric power system. There are many factors which affect electrical load forecasting. Therefore, a hybrid model is required to improve the forecast and make it more accurate. This paper presents a new hybrid load estimation method using two approaches: Wavelet transform (WT) and Artificial neural network (ANN). In order to take into account the large asymmetric time-varying electric raw data set, wavelet technique is used to decompose the data in terms of both time and frequency. Several wavelet functions are available, but selecting a proper wavelet function plays a crucial role in designing the model. In present work, the following types of wavelet functions, namely Haar, Deubechies, Symlet, Coiflet have been used to disintegrate the electrical load data into distinct segments. Later, ANN has been employed to forecast the non-linear data of the load. The proposed model is validated through AEMO data for 24 hours of a day over a one-week period.