A data-driven feed-forward decision framework for building clusters operation under uncertainty

• A self-tuned and noise-tolerant building clusters model is developed.
• A feed-forward decision framework is proposed for building clusters operation.
• Four data fusion techniques are compared in terms of accuracy and robustness.
• Operation decision obtained by the proposed framework can achieve more cost saving.

Building plays a significant role for energy consumption and carbon dioxide emission in the United States. Extensive researches are conducted to develop effective operation strategy for the building system. However, less study is to investigate the energy sharing among a cluster of multiple buildings (aka building clusters) under uncertainty. In this research, we propose to develop a data-driven feed-forward decision framework for building clusters operation, through the use of noise-tolerant data fusion techniques. Three stages are implemented in the proposed framework which include: (1) decisions generation stage that employs an augmented multi-objective particle swarm optimization based decision framework to obtain operation decisions for the next future L hours; (2) execution stage that implements the first l hours decisions; and (3) calibration stage that employs data fusion techniques to calibrate the building clusters model in a l′ hour scale. The calibrated model is fed back to the decisions generation stage for the next period decisions. Unscented Kalman filter which is demonstrated to outperform other data fusion techniques in terms of accuracy, robustness and computational efficiency based on our experimental results is employed in the calibration stage. To evaluate the performance of the proposed framework, we compare the operation decisions with and without calibration stage. It is demonstrated that the proposed feed-forward framework can obtain operation decisions to achieve more cost savings. The impacts of different time lengths l in the execution stage are investigated which indicate the selection of l depends on the trade-offs between decision solution quality and computational performance.