Short-term Meter Level Load Forecasting of Residential Customers Based on Smart Meter's Data

Decentralization of conventional power system into small scale, smart and efficient micro-grids emphasizes the requirement of short-term load forecasting on the minute level. The postulation of short-term load forecasting for household level has evolved with this transition in the power system. However, fluctuations and uncertainty in the load profile of individual customers make it difficult to predict. This paper presents the implementation and comparison of three machine learning based approaches to accurately forecast the meter level aggregate demand using consumption data of a few major appliances. The forecasting techniques include the feed-forward neural network (FNN), long short-term memory (LSTM) network and particle swarm optimization (PSO) based FNN. The results prove that PSO-FNN exhibits better forecasting accuracy while the conventional FNN has better computational efficiency.