Electrical energy plays a vital role in daily life. Since the adoption in 2015 of the new United Nations (UN) Sustainable Development Goals (SDGs), a lot of attention has been paid to the energy sector in particular. The SGD number 7 aims to ensure by 2030, access to affordable, reliable, sustainable, and modern energy for all. Thus, understanding electricity consumption growth remains a fundamental aspect of achieving universal access to energy. This research aimed to investigate the multivariate modelling of yearly residual electricity consumption in Togo by conducting a comparative approach between different deep learning algorithms. The analysis uses a data-driven approach based on postpaid residential electricity consumption data from Togo Electricity Company, aggregated at the regional level, from 1994 to 2020. Alongside this consumption data, additional covariates were selected through a feature selection procedure. The data were modelled using deep learning algorithms. Three algorithms were used namely Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model. The model’s performances were assessed using the Mean Absolute Error (MAE) metric. As a result, from this study, the CNN-LSTM hybrid model performed the best with an MAE of 6.2 GWh on the test set while MLP, CNN, and LSTM performed with an MAE of 9.1 GWh, 14.4 GWh, and 14.8 GWh respectively. These results revealed the potential of deep learning techniques in modelling yearly residential electricity consumption in Togo. Thus, the utility company can rely on a such deep learning model to right-size its residential electricity supply for a more sustainable energy use and a fair and cost-efficient green energy transition.