RNN-Based Deep Learning for One-hour ahead Load Forecasting

Accurate load forecasting plays a key role in economical use of energy. Artificial Neural Network (ANN) models have been extensively implemented to produce accurate results for short-term load forecasting with time lead ranging from an hour to a week. In this report daily peak load forecasting has been performed for the part of a town supplied by 2 distribution feeders on weekdays by taking into consideration the historical maximum Power consumption in MWH, Voltage in KV and Current in Amp data. Optimization of the network parameters is performed for both learning rules. Energy demand forecasting is of great importance in the management of power systems. In this report artificial neural network technique (ANN) is used for forecasting the load curve. Algorithms using these techniques have been programmed using deep learning case study. The efficiency of both the model is determined from the load curve and the load is predicted as a testing sample. The Recurrent Neural Network model trains the daily load data for a set of days and then forecast the load for next day. Actual data are obtained from "Mankapur Substation" is used to validate the result.

Keywords: Machine Learning, Output Load Forecasting, Recurrent Neural Network.