Machine learning based renewable energy generation and energy consumption forecasting. Issue 2 (1st June 2023)

• Record Type: Journal Article
• Authors: Talwariya, Akash; Singh, Pushpendra; Jobanputra, Jalpa H.; Kolhe, Mohan Lal
• Is Part Of: Energy sources. Volume 45:Issue 2(2023)
• Page Start: 3266
• Page End: 3278
• Publication Date: 2023-06-01
• Subjects: Renewable energy forecasting -- Energy demand forecasting -- Solar energy forecasting -- Wind energy forecasting -- Machine learning; Natural resources -- Periodicals; Energy consumption -- Periodicals; Energy consumption -- Climatic factors -- Periodicals; Energy conversion -- Periodicals; Energy conversion -- Environment aspects -- Periodicals; Power (Mechanics) -- Periodicals; 333.7905
• DOI: 10.1080/15567036.2023.2194257
• Languages: English
• ISSNs: 1556-7036

Abstract

ABSTRACT: Renewable energy generation sources are the need of the hour to mitigate the challenges of fossil fuels, environmental impacts, large transmission and distribution losses, and variable consumption pattern. Renewable energy such as solar PV generation or wind generation is most unpredicted generation and depends upon weather constraints. A promising prediction of generation is a suitable solution for this problem. A machine learning-based conventional neural network algorithm is proposed to forecast the generation and calculate the errors in generation forecast. The generated energy is supplied to consumers, but the consumers consume electricity by their choice and an energy management model is required at the consumer end to supply the available predicted generation to develop a self-sustainable microgrid. The energy consumption at consumer end is essential to predict, and based on the predicted consumption at hourly bases, the cost model for the consumers is developed. The consumption profile is forecasted using long short-term memory as it is a good tool for time-series forecasting using available sequential load consumption data. The demand is forecasted and analyzed for selected load categories (i.e. fixed, non-shift-able, and shift-able loads). The evaluation matrix mean square error (MSE), mean absolute error, and root mean square error (RMSE) are used for the error calculations. The RMSE for solar and wind power generation forecasting have been 2.472 and 3.034, respectively. The RMSE for fixed, non-shiftable, and shiftable demand forecasting have been 0.027, 0.067, and 0.21, respectively. The results show that the proposed methods for predictions of solar, wind power generation and energy consumption forecast achieve better accuracy. The presented methodology can be implemented with energy storage for renewable energy production and demand management for energy from sustainable energy resources.