Solar PV Power Forecasting Using pvlib Physics Model and Machine Learning

A grid-connected solar plant has to declare its output for the next day, and its owner wants to know whether the plant performs as expected. This project forecasts the hourly AC power of a real 52.9 kWp rooftop PV plant (NREL PVDAQ system 1199, Hunt Valley, Maryland, USA) from weather data. It compares the classical approaches - smart persistence (the clear-sky index of yesterday) and the pvlib PVWatts physics model (solar position, clear sky, Perez transposition, SAPM cell temperature, PVWatts DC and inverter) plus an independent PVGIS simulation - with Random Forest, LightGBM gradient boosting and an MLP neural network that use the weather and the pvlib outputs as features (hybrid physics + machine learning). The models are fitted on 2016, chosen on 2017 and tested once on 2018. An 80 % prediction interval, error analysis by sky class, a web application with five databases, a GNU Octave cross-check (run) and a complete MATLAB version (written, not run) are included. Everything runs on a student laptop CPU in a few minutes.

Category: Electrical and Electronics Engineering (Renewable Energy, Solar Photovoltaics, PV System Modelling, Power-System Operation, Machine Learning, pvlib, PVGIS, MATLAB, GNU Octave, Python)

Electrical & Electronics Engineering Project EE02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.