Methodological framework for short-and medium-term energy, solar and wind power forecasting with stochastic-based machine learning approach to monetary and energy policy applications

Anomalous seasons such as low-wind summers and extremely cold winters can seriously disrupt energy reliability and productivity. Better short/medium- term forecasts that provide reliable and strategic planning insights will allow the energy industry to plan for these extremes. In order to efficiently quantify uncertainty, this study proposes a Gaussian stochastic-based machine learning process model (GPR) for short/medium-term energy, solar, and wind (ESW) power forecasts using two different temporal resolutions of data. Four experimental steps (EXMS) were designed. Each EXMS is designed with four distinct fitting and predicting methods, and the GPR model uses seven kernel covariance functions for hyperparameter optimization. Real-time data is used for the forecasting analysis at three different locations. The forecasting results are validated using three existing models. The percent coefficient of variation of CVGPR1 and CVGPR2 of EXMS-1 and EXMS-3 for ESW power forecasts is 0.017%, 0.057%, 0.025%, and 0.223%, 0.225%, 0.170%, respectively. Accuracy has shown that the proposed model can predict ESW power simultaneously at two different temporal resolution data. The GPR accuracy with four EXMS methodologies is promising by addressing ESW power forecasts under the GPR framework of significant utilities, independent power producers, and public interest. (C) 2021 Elsevier Ltd. All rights reserved.

Automated summary

The study proposes a Gaussian stochastic-based machine learning process model for short/medium-term energy, solar, and wind power forecasts, demonstrating promising accuracy and reliability through multiple experimental steps and various methods for validation.