Prévision de la demande en énergie électrique au Cameroun par les méthodes de régressions et des reseaux de neurone artificiel

Abstract

The main ob jective of this thesis is to build models for forecasting and efficient management of the electrical energy available in Cameroon. With this in mind, we have proposed optimal numerical models based on hybrid methods of artificial intelligence and methods of econometrics. For the long-term forecasts, the vec- tor error correction model (VECM), the exponential smoothing of Hol (HES), the fuzzy logic (FLS) and the hybrid model (VECM-HES) were used. In the context of short-term forecasting, multiple linear regression (MLR) models, Hol exponential smoothing (HES), artificial neural networks (ANN), deep neural networks (CNN and DNN) as well as the hybrid model (ANN-LRM-HES) were used. A compa- rative study of the different results obtained by the hybrid models VECM-HES and ANN-LRM-HES show that these models display a relatively low margin of error, compared to the margins of error obtained by similar models existing in the literature on the sub ject. We therefore conclude that these proposed hybrid models are better suited for our forecasting calculations. In continuation for the long-term forecasts, the results obtained also allow us to conclude that in the same way as the evolution of the population, the consumption of electric energy has an increasing tendency. According to our forecasts, it goes from 7169,031 GWh in 2020 to 8134,772 GWh in 2024, an increase of 965,741 GWh in five years.