This project investigates the use of deep learning models to forecast the household power consumption of residential households in the UK. For the raw data, we used energy consumption data for 5,567 households in London obtained from the Low Carbon London Project. We performed dimension reduction and clustering to find representative households for training. Naive and Seasonal ARIMA models were used as baselines. Using MSE as a metric, both models offer impressive performance on short-term forecasting, but the performance drops dramatically when considering long-term forecasting. To address this drawback, we trained and compared two sequence-to-sequence LSTM forecasting models with different household selections. Overall, the model captures trends well in most of the time period and demonstrates good generalization, even for unseen households.