PREDICTION OF SOLAR ENERGY GENETRATION FROM THE WEATHER DATA USING MACHINE LEARNING

Abstract

The significant growth of the amount of grid energy supplied by renewable sources is one of the main objectives of smart grid efforts. One of the challenges in incorporating renewable energy sources into the system is the intermittent and unpredictable nature of electricity generation. The necessity to relocate generators to meet demand as production fluctuates makes it imperative to forecast future renewable energy output. While building complex prediction models by hand for huge solar farms may be feasible, doing so for distributed power generation in the grid's millions of homes is a difficult undertaking. This research investigates machine learning methods for automatically generating site-specific forecasting models for solar power generation using National Weather Service (NWS) weather predictions in order to address the problem. by comparing several regression techniques to create prediction models, including multilayer perceptron’s and neural networks with long-term memory. combining historical NWS forecasts and sun intensity data from a weather station that has been operational for about a year to assess the accuracy of each model. Our findings demonstrate that predictive models developed for our site employing seven different weather forecasting parameters are more accurate than current forecast-based models.