The aim of STLF is to predict the future electricity load based on the recognition of similar repeating trends of patterns from historical load data. Normally, there are two different methods of forecasting models - the traditional models and the modern techniques. Traditional forecast models entail time series and regression analysis through the use of statistical models such as exponential smoothing, regression, Box-Jenkins model, State-Space Model, Kalman Filter etc. These models are mostly linear methods and have limited ability to capture non-linearities in the load time series pattern. They require expert knowledge and are much complex to operate. In the recent years, many researchers switched to try the modern techniques based on artificial intelligence. Of all, the Artificial Neural Network (NN) receives the most attention. NN is regarded as an effective approach and is now commonly used for electricity load forecast. The reason for its popularity is its ease of use and its ability to learn complex input-output relationship. The ability to learn gives NN a better performance in capturing nonlinearities for a time series signal. Therefore, this project proposes a model comprising neural networks as its forecasting tool. A NN based forecast model requires historical electricity load data as the input variable to perform STLF. It contains some hidden time-variant pattern. These patterns could be informative to a NN based forecaster to enhance the NN’s ability in learning the signals. Thus to extract hidden patterns from the electricity load data, a wavelet decomposition technique is introduced. In the proposed model, initially wavelet decomposition has been performed on historical electricity load data, historical previous day data and historical price data of one month. Target data for next year same month also used for the training of Neural Network model. The useful decomposed components of both based on energy level is then feed to the NN model for training, testing and validation. Finally the output of NN model is converted into wavelet reconstruction which is the predicted data.
Loading....