Load forecasting plays a major role in determining the prices of the energy supplied to end customers. An accurate prediction is vital for the energy companies, especially when it comes to the baseline calculations that are used to predict the energy load. In this paper, an accurate short-term prediction using the Exponentially Weighted Extended Recursive Least Square (EWE-RLS) algorithm based upon a standard Kalman filter is implemented to predict the energy load for blocks of buildings in a large-scale for four different European pilot sites. A new software tool, namely Local Energy Manager (LEM), is developed to implement the RLS algorithm and predict the forecast for energy demand a day ahead with a regular meter frequency of a quarter of an hour. The EWE-RLS algorithm is used to develop the LEM in demand response for blocks of buildings (DR-BOB), this is part of a large-scale H2020 EU project with the aim to generate the energy baselines during and after running demand response (DR) events. This is achieved in order to evaluate and measure the energy reduction as compared with historical data to demonstrate the environmental and economic benefits of DR. The energy baselines are generated based on different market scenarios, different temperature, and energy meter files with three different levels of asset, building, and a whole pilot site level. The prediction results obtained from the Mean Absolute Percentage Error (MAPE) offer a 5.1% high degree of accuracy and stability at a UK pilot site level compared to the asset and whole building scenarios, where it shows a very acceptable prediction accuracy of 10.7% and 19.6% respectively.