Short Term Load Forecasting Using Artificial Neural Networks With Supply-Demand Based Hourly Information
Electricity load forecasting is an important risk management and planning tool for electric utilities as a conception of forecasting. Forecasting process may be summarized as calculation of the probability of prudential circumstance with analyzing retrospective data. Load forecasting helps producers and official planners to make important decisions including purchasing and generating electric power, load switching, and infrastructure development. This study proposes an artificial intelligence model with using publicly accessible historical electricity load data from Turkey Electricity Transmission Joint-Stock Corporation official website for the following dates between January 2010 – November 2012. The proposed model used system marginal price (SMP), market proffer price (MPP), calendar effect and hourly electricity load data as input and finalized system electricity load as output. Artificial intelligence models using Multilayer Perceptron are analyzed with Levenberg-Marquardt Algorithm (LMA), Fletcher-Powell (CFG), Quasi-Newton (BFGS) back-propagation algorithms and performed maximum classification % 91.325 succeed with 0,00188 training error.
Keywords— Load forecasting, artificial neural networks, Levenberg-Marquardt (LMA), Fletcher-Powell (CFG), Quasi-Newton (BFG).