A review on renewable energy and electricity requirement forecasting models for smart grid and buildings

The benefits of renewable energy are that it is sustainable and is low in environmental pollution. Growing load requirement, global warming, and energy crisis need energy-intensive management to give sincere attempts to promote high accuracy energy monitoring techniques in order to enhance energy system efficiency and performance. The energy consumption data of domestic, commercial and industrial are becoming accessible to estimate the notable share of various sectors in the energy market. Energy forecasting algorithms play a vital role in energy sector development and policy formulation. Energy prediction and power supply management are the key roots of energy planning. A large number of prediction models have been used in the recent past. The selection of a prediction model usually based on available data, the objectives of the model network mechanism and energy planning operation. In this review, we conduct a critical and systematic review of renewable energy and electricity prediction models applied as an energy planning tool. The forecasting intervals is divided into three sections including: i) short-term; ii) medium-term; iii) and long-term. Three renewable energy resources, i.e. wind, solar, and geothermal energy, and electricity load demand requirement are considered for review forecasting analysis. Three major states-of-art forecasting classifications: i) machine learning algorithms; ii) ensemble-based approaches; iii) and artificial neural networks are analyzed. These approaches are investigated for prediction applicability; accuracy for spatial and temporal forecasting; and relevance to policy and planning objectives. The machine learning models can handle large amount of data with accurate forecasting analysis. Applying ensemble techniques enables us to obtain higher forecasting accuracy by combining different models. Artificial neural networks if used in the right way can contribute a robust choice, given that it is capible to extract and model unseen relationships and features. Furthermore, unlike these conventional techniques, artificial neural networks do not force any limitation on residual and input distributions. Findings from this review would help professionals and researchers in obtaining recognition of the prediction approaches and allow them to choose the relevant methods to satisfy their desired tasks and forecasting requirements.