Research paper Forecasting gas energy use presented at CLIMA 2019
- 13 juni 2019
- Posted by: S. Eijlander
- Category: nieuws
Baldiri Salcedo presented last week at the CLIMA 2019 in Bucharest a research paper about forecasting energy use produced in collaboration with a group of students of the minor Applied Data Science: Brian de Keijzer, Pol de Visser, Víctor García Romillo, Víctor Gómez Muñoz, Daan Boesten and Megan Meezen. The aim of the research was to test the different machine learning algorithms on forecasting gas use making use only of weather data and time related data. The algorithms were tested on a gas use dataset from the OPSCHALER project.
The paper comes to the conclusion that with this amount of data simple forecasting methods like Multivariate Linear Regression score as good as more complex machine learning algorithms as Deep Neural Networks, Recurrent Neural Networks or Convolutional Neural Networks. However, complex methods should obtain considerable better forecast if more data about the houses was given to the algorithms. Especially interesting was the use of Convolutional Neural Networks, usually use for image recognition, by transforming the data first into an image.