/Power-Generation-Prediction

A Time-series forecasting case study

Primary LanguageJupyter NotebookMIT LicenseMIT

Solar and Wind Power Generation Forecasting

A Time-series forecasting case study

Last Update: June 8, 2022

N|Solid

Integration of solar and wind energy sources with the grid requires predicting those sources since real-time computation is not feasible for power system planning. Furthermore, integration with the electricity market is another matter that makes generation forecasting vital. This research aims to investigate various machine learning methods for forecasting solar and wind power generation. Using two datasets containing the meteorological and generation history of the state of California from 2018 to 2020 in an hourly manner were utilized for training four regression and three neural networks (NN) models for predictions of the solar and wind generation. NN techniques were found to forecast with the least error. Also, the fastest and slowest methods in terms of training speed were identified. This study answers the question regarding the prediction of solar and wind power generation using meteorological data from renewable generation plants. A model trained with a geographically accurate dataset might be used to plan ahead solar and wind power generation to improve grid integration and the energy market.

Methods

  • Linear Regression
  • SVM Regression
  • GBoost
  • XGBoost
  • CNN
  • LSTM
  • ✨ GRU✨