Session: 17-01-01: Research Posters

Paper Number: 149640

149640 - Identifying the Impact of Metrological Parameters for Energy Demand Prediction in the Usa Using a Hybrid Svm-Pso Optimization Technique 

The growth of the economy is heavily dependent on the consumption of energy, which negatively contributes to the global warming, glacier melting, fertile land degradation, and food shortages. To alleviate these adverse effects, it is crucial that we transition towards renewable resources. Despite having lower solar radiation conversion efficiency, photovoltaic (PV) panels offer valuable benefits. A wide range of factors can influence the performance of a PV panel, including solar radiation, temperature, wind speed, relative humidity, location, incident angle, shading, and dust accumulation.The aim of the present research is to determine the most impactful meteorological parameters affecting PV panel output, and subsequently use that knowledge to forecast the power generation output in the city of New York, USA. The input dataset comprises direct solar radiation, diffuse solar radiation, clearness index, maximum, minimum, and average temperatures, wind speed, and relative humidity, all obtained from the Data Access Viewer, National Aeronautics and Space Administration (NASA), whereas the Solar photovoltaic power plant data is obtained from National Renewable Energy (NREL) Laboratory. The dataset has been scaled using a robust scaler. A Gradient Boosting algorithm with Bayesian hyperparameter tuning, along with K-fold cross-validation was employed to identify the most influential parameters. Additionally, a Support Vector Machine with metaheuristic optimization i.e. Particle Swarm Optimization was used to predict the PV panel power output. The results reveal that diffuse horizontal irradiance, direct normal irradiance, wind speed, and maximum temperature are the most influential parameters, with feature importance scores of 0.27, 0.22, 0.17, and 0.13, respectively. The best loss value obtained during the hyperparameter tuning using the Bayesian optimization is 2.97 indicating the mean squared error (MSE) of the model on the cross-validation fold. Furthermore, the prediction of PV panel power output achieved accuracy with mean absolute error, mean percentage error, and coefficient of determination reported as 1.26, 3.24%, and 0.91, respectively. The best parameters for hyper parameter tuning are a Learning rate 0.031, estimators 75, maximum depth 4 and, minimum sample split of 5. Thus, solar irradiance is the most influential parameter driving power generation, followed by wind speed, which provides passive cooling. Maximum temperature decreases the efficiency of the PV panel power output. Overall, the findings identify key meteorological factors that significantly impact power output and utilize them to improve forecasting accuracy. By gaining a deeper understanding of these influences, policymakers and energy managers can optimize power output, enhance resource allocation, and strengthen energy infrastructure resiliency.

Presenting Author: Rahul Makade Stevens Institute of Technology

Presenting Author Biography: Dr. Rahul Makade is a postdoctoral researcher in Energy, Control, and Optimization (ECOLab) at Stevens Institute of Technology.

Authors:

Rahul Makade Stevens Institute of Technology
Gizem Acar Stevens Institute of Technology