The purpose of this project is to forecast DA solar energy production at Oklahoma Mesonet stations using regression machine learning algorithms with the numerical weather prediction (NWP) model data from NOAA/ESRL Global Ensemble Forecasting System (GEFS) as input. Accurate forecasting of solar energy generation is very important in order to optimize the balance between energy production from fossil fuel and solar plants. It helps operators make better day ahead unit commitment decisions which help to balance the power reserve and reduce the power cost. This project was initially organized by the American Meteorological Society and posted as a Kaggle contest. The complete dataset required for this project can be obtained from the kaggle site page. The entire project is presented in 3 major parts (nootbooks). Python 3.6.4 version has been used for coding.

1. Data wrangling: Data preparation for exploratory analysis and modeling

2. Data story: Exploratory analysis to examine underlying structures, detect outliers and anomalies, uncover patterns, find correlations, test underlying assumptions etc. The data presented in this section is only from 'ACME' station, Oklahoma, USA.

3. Model prediction: Different machine learning models are employed to predict solar energy using weather variables as input features. The data presented in this section is only from 'ACME' station, Oklahoma, USA.

End to end coding: This notebook contains end-to-end coding for all 98 mesonet stations in Oklahoma, USA. Although it was a late submisison my scoring holds 79th rank in the contest.

Full project report is avaibale here.

Presentation slides for this project is avaiable here.