Short Term Load Prediction in Distributed System Using Machine Learning

Accurate forecasts of electrical substations are mandatory for the efficiency of the Advanced Distribution Automation functions in distribution systems. Artificial Neural Network (ANN) model have been extensively implemented to produce accurate results for short-term load forecasting with time lead ranging from an hour to a week. The proposed system describes the design of a class of machine-learning models, namely neural networks, for the load forecasts of medium-voltage/low-voltage substations. Four weather seasons are defined by the Meteorological Department, India. Each season includes the group of month. Representative months are selected from each season by observing the variation in load behavior patterns. An input vector composed of load and temperature values at previous instants, is employed to train ANN designed for each selected month by using Back-Propagation algorithm with Momentum learning rule. We focus on the methodology of neural network model design in order to obtain a model that has the best achievable predictive ability given the available data. ANN testing is carried out and their performance is evaluated using mean absolute percentage error (MAPE) criterion. Finally, error values are compared for each month and hence the deviation in forecasting ability of ANN is observed for each month and season.