Parameter Tuning of Deep Learning Using Evolutionary Algorithm

Défense de mémoire de Maxime Hendrix

The deep learning is an algorithm based on machine learning that allows to forecast the consumption of electricity in Spain. The deep learning algorithm receives input data that represent the past consumption of electricity. After dierent computations, the
desired forecast is obtained as an output. However, the deep learning algorithm has some parameters that need to be congured in order to predict with accuracy. A Genetic algorithm is used to perform this parametrization and to nd the optimal parameters.
The goal of this thesis is to optimize the parameters of deep learning algorithm in order to forecast with more accuracy the consumption of electricity in Spain. The results obtained by the deep learning algorithm with an optimization of the parameters are better
than without. There are other methods that allow to forecast the electricity consumption. The comparison between our developed algorithm and the other techniques shows that our algorithm is more ecient to forecast the consumption of electricity.

keywords: machine learning, genetic algorithm, deep learning algorithm, forecasting, time series, optimization parameters

 

Contact : Isabelle Daelman - 4966 - isabelle.daelman@unamur.be
Télecharger : vCal