M. H. M. R. S. Dilhani, N. M. Wagarachchi, K. J. C. Kumara
Abstract
Due to the rapid increment of daily energy demand in the world, electric load forecasting becomes one of the most critical factors for the economic operation of power systems. The energy usage of the electricity demand is higher than the other energy sources in Sri Lanka (according to the record of Generation Expansion Plan – 2016, Ceylon Electricity Board Sri Lanka). Moreover, the forecasting is a hard challenge due to its complex nature of consumption. In this research, the long term electric load forecasting based on Optimized Artificial Neural Networks (OANNs) is implemented using Particle Swarm Optimization (PSO) and results are compared with a regression model. Results are validated using the data collected from central bank annual reports for thirteen years from year 2004 to 2016. The choice of the inputs for ANN, OANNs and regression models are based on the results gained from the correlation matrix. In this process, all the training data sets are scaled to be between 0 and 1. To meet this purpose, the data set is divided by its largest value. Moreover, the results show that the accuracy of forecasting using OANN is better when it is compared with ANN and regression model. Forecasting accuracy of each model is performed by Mean Absolute Percentage Error (MAPE).
Keywords: Backpropagation, Electricity Load Forecasting, Neural Network, Particle Swarm Optimization, System-type Architecture
**View Full Paper in the Springer Source