Applying Machine Learning Methods for Power Plant Generation Time Series Forecasting

Abstract

The paper highlights an approach to predicting the generation of a thermal power plant using machine learning methods. In the course of the work, features were generated based on electrical and date-time values, and modeling was carried out using two architectures of recurrent neural networks at the first stage and three-level ensembles of models were built based on linear regression and gradient boosting over decision trees at the second stage. The obtained quality metrics make it possible to judge the fundamental possibility of using the considered method for solving both this and related problems related to forecasting time series.