Abstract
The modern power grid faces challenges regarding many complex factors affecting both demand and generation; including growth in demand; incorporating large-scale renewable power penetration; uncertainties in climate change; lack of historical data; and coordination of the large volume of data. These issues have resulted in complications in forecasting load and generation in micro grids. The loads are becoming more erratic and the generation is intermittent. Thus, this paper presents a study of different forecasting approaches for load and generation, by comparing multiple univariate and multivariate methods to analyse their effect. The study also proposes seasonal models: the SARIMA model taking into consideration the historical load, the correlation of weather data and renewable integration to estimate future behaviour of the microgrid by predicting one day ahead using critical load data; whereas the Holt Winters method is used for generation forecasting. A case study is simulated using real-world microgrid data for the selected geographic location in Australia. The results suggest that for the day-ahead load forecast, the SARIMA model performed relatively better compared to MLR, Holt-Winters additive and multiplicative methods; whereas, for generation forecasting, Holt-Winters Additive Method and SARIMA perform well for Autumn and Summer respectively. The results suggest that the proposed approach of using different seasonal models for load and generation forecasting yields higher accuracy as compared to conventional forecasting.