Data-driven thermal dynamics recognition and multi-objective optimization for building demand response

Abstract

Large-scale integration and management of air-conditioning (AC) systems is crucial to the decarbonization of electric power systems. The current challenges lie in the computational accuracy and speed of thermal dynamics recognition and optimization, and there is still a lack of effective methods to deal with the situation of complex indoor spaces and various AC systems. In addition, users often have diverse preferences between operational costs and thermal comfort under different conditions, which are vague for optimal operation. Hereby, this paper introduces a physics-restricted state-space (PRSS) building thermal dynamics recognition method and a rolling-horizon optimization approach. The recognition method utilizes a high-dimensional linear regression to calibrate the state-space equation with physical constraints. An adaptive weights algorithm is also introduced to extract the historical preference between thermal comfort and operational cost, and determine a range of weights for better performances in cost and thermal comfort than the history operation. Long-term operational data of a commercial building (38 indoor zones) and an office building (8 indoor zones) with different AC systems are utilized to validate the proposed method. The proposed recognition method can get predictions of multi-zone air temperature with the rooted mean squared error and the mean absolute error both less than 0.3 K. Moreover, the proposed optimization approach can fulfill the multifaceted requirements of demand response, including immediacy for incentive-based programs and reliability for price-based programs. Our work demonstrates its potential to be a generic interface to facilitate large-scale regulation of AC systems.