The main purpose of load forecasting is to provide an accurate estimation of the future electricity demands, which is becoming increasingly important for the operation and planning of existing electricity network and future smart grids, as they will feature much higher ranges of uncertainties, larger variations of power flows and increased levels of interactions between supply and demand sides. Typical load profiles exhibit periodicity, allowing to extract patterns from demand time series and available historical recordings. However, there are many factors that cause strong variations of these demand patterns, including calendar and socio-behavioral aspects (time of the day, day of the week, season of the year, but also weekly working schedule, public holidays, etc.), as well as meteorological or weather related factors (ambient temperature, solar irradiation, precipitation, etc.). This paper analyzes load forecasting using a stacked bidirectional long short-term memory (SB-LSTM) recurrent neural network based approach, which is a state-of-the-art method for regression analysis of time-series data under deep learning framework. The analysis is performed on a case study of residential demands in Scotland, for which a five-year-Iength datasets containing both load and weather data recordings are available. The presented results and analysis allow to evaluate how accurately SB-LSTM approach can provide predictions for both day-ahead and week-ahead load forecasting, taking into account meteorological information.