Finding Power: Utilizing a Predictive Neural Network to Find Potential Solar Energy Outputs

Climate change needs to be addressed. The warming of the Earth has caused a significant amount of heat-induced deaths, raised the sea level, and created more devastating tropical storms. These effects will worsen if the climate continues to get warmer at its current rate, endangering the lives of humans across the globe. The burning of fossil fuels for energy produces billions of tons of greenhouse gases and we need to switch to a cleaner energy source, solar panels. However, misconceptions about solar energy prevent an effective transition from taking. To address these assumptions, we created a neural network to predict solar energy outputs so that consumers have an accurate estimate of how much energy they may receive and budget their use of it accordingly, creating confidence in solar energy. Using the 2021 Photovoltaic Voltaic Power Output Dataset, we trained and evaluated our neural network model so that it understood how important variables (global irradiance, direct irradiance, temperature, etc.) impacted solar panel output. At the end of the model’s training, it achieved a SMAPE (symmetric mean absolute percentage error) of 12%. We intend to collect more data to improve the model’s SMAPE to provide even more accurate estimates people can rely on. In addition, we seek to expand on our model and create an active energy counter that tracks how much energy one has available to them and takes into account the use of appliances and their energy demand to help people effectively budget their energy and feel more secure in renewable energy sources.