Power systems are undergoing a significant shift towards renewable energy technologies. To make the most of these energy sources, optimizing the generation, storage and distribution of energy can be enhanced with information about future energy consumption. Forecasting the consumption load of individual residents plays a key role in load balancing but is challenging due to the irregular nature of individual consumption patterns. Further, current literature is limited to forecasting residential load to a few hours in the future. In this paper, we propose Fusion ConvLSTM-Net, a fusion encoder-decoder architecture that combines spatial and temporal features to extend the load forecast to a full 24 hours. We evaluate the model against other benchmark neural network models across various forecast window sizes and average model performance over multiple houses. Additionally, we conduct analyses on its generated forecasts to detect model degradation, which often occurs with a shorter prediction window size. Our extensive experiments demonstrate that the Fusion ConvLSTMNet significantly increases the forecast window, prevents model degradation and delivers the most accurate performance as compared to other benchmark neural networks.