Thermal storage systems were originally designed to shift on-peak cooling production to off-peak cooling production in order to reduce on-peak electricity demand. Recently, however, the reduction of both on- and off-peak demand is a critical issue. Reduction of on- and off-peak demand can also extend the life span and defer or eliminate the replacement of power transformers. Next day electricity consumption is a critical set point to operate chillers and associated pumps at the appropriate time. In this paper, a data evaluation process using the annual daily average cooling consumption of a building was conducted. Three real-time building load forecasting models were investigated: a first-order autoregressive model (AR(1)), an autogressive integrated moving average model (ARIMA(0,1,0)), and a linear regression model. A comparison of results shows that the AR(1) and ARIMA(0,1,0) models provide superior results to the linear regression model, except that the AR(1) model has a few unacceptable spikes. A complete control algorithm integrated with a corrected AR(1) forecast model for a chiller plant including chillers, thermal storage system, and pumping systems was developed and implemented to verify the feasibility of applying this algorithm in the building automation system. Application results are also introduced in the paper.