Abstract
A pertinent problem of renewable energy source integration to main utility power, requiring a smart grid is brought up and its solutions are investigated from a data scientist’s perspective. A smart grid design aims for renewable energy integration with traditional electricity supply with a more efficient, data-driven autonomous control mechanism for better quality service at an optimal cost to end-users. This digital transformation of the power system needs the reskilling of software engineers as data scientists. From a data scientist’s perspective, it is a problem of efficiently generating accurate forecasting of energy consumption of a household or a larger region from past observations for uninterrupted power supply at optimal cost. In this paper, the traditional statistical method of autoregressive integrated moving average model for time series (TS) data analysis and forecasting is discussed first. Then, for more accurate forecasting for nonlinear TS data, deep learning approach of artificial neural network which are multilayer perceptron and long short-term memory recurrent neural network models are discussed and experimented with. To support the discussion, performance metrics for all three algorithms are calculated for four-TS datasets of varied length and structure and compared in terms of their salient features, complexity and accuracy.
