Peak power shaving is a technique that can be used to reduce monthly electricity bills. As control of Energy Storage Systems (ESS) is based on predicted power demand, power demand forecasting is a necessary component of entire building power optimization. Various forecasting methods have been developed. However, the importance of intra-day prediction error is overlooked by present models. In this paper, a variety of dynamic intra-day model modification strategies utilizing intra-day prediction error are proposed to improve power demand prediction and peak shaving performance. These modification strategies could be applied to any models which do the prediction at the beginning of the day. A Self-Organizing Map (SOM) & Support Vector Regression (SVR) Adaptive Hybrid Model proposed in previous literature is chosen as baseline in this paper. The method of bisection is adopted to calculate the optimal threshold to control the ESS. Simulation results demonstrate effectiveness of intra-day prediction modification strategies.

This content is only available via PDF.
You do not currently have access to this content.