Privacy-Preserved Aggregate Thermal Dynamic Model of Buildings

TL;DR

This paper tackles the privacy risk in estimating the aggregate thermal dynamic model ($ATDM$) for building clusters under direct load control. It introduces a privacy-preserved parameter estimation framework that combines transformation-based encryption ($TE$), secure aggregation protocol ($SAP$), and block coordinate descent ($BCD$) to infer ATDM parameters without exposing individual indoor temperatures or loads. The method provides theoretical privacy guarantees by reducing privacy leaks to solving a multivariate quadratic system (MQS), which is NP-hard, and demonstrates practical effectiveness on real data with only minor accuracy loss compared to non-private methods. Simulation results show accurate aggregate-state predictions, effective data masking, and acceptable computational overhead, highlighting the framework’s potential for privacy-aware demand response and grid operations. Overall, the approach enables building clusters to participate in energy systems with preserved privacy while maintaining high-quality aggregate dynamic models.

Abstract

The thermal inertia of buildings brings considerable flexibility to the heating and cooling load, which is known to be a promising demand response resource. The aggregate model that can describe the thermal dynamics of the building cluster is an important interference for energy systems to exploit its intrinsic thermal inertia. However, the private information of users, such as the indoor temperature and heating/cooling power, needs to be collected in the parameter estimation procedure to obtain the aggregate model, causing severe privacy concerns. In light of this, we propose a novel privacy-preserved parameter estimation approach to infer the aggregate model for the thermal dynamics of the building cluster for the first time. Using it, the parameters of the aggregate thermal dynamic model (ATDM) can be obtained by the load aggregator without accessing the individual's privacy information. More specifically, this method not only exploits the block coordinate descent (BCD) method to resolve its non-convexity in the estimation but investigates the transformation-based encryption (TE) associated with its secure aggregation protocol (SAP) techniques to realize privacy-preserved computation. Its capability of preserving privacy is also theoretically proven. Finally, simulation results using real-world data demonstrate the accuracy and privacy-preserved performance of our proposed method.

Figures (6)

• Figure 1: Concept of the Aggregation of the building cluster.
• Figure 2: Information exchange of the proposed algorithm. The blue arrows between the BLA block and the building cluster block represent the interactive information within these two entities. The blue arrows within the building cluster block denote the intermediate information between two agents that is necessary for SAP. The red arrows represent the algorithm procedures performed between information transformations, among which the first two belong to $\mathrm{SP}_\mathrm{I}(\xi)$ and the others belong to $\mathrm{SP}_\mathrm{II}(\alpha)$.
• Figure 3: Aggregation coefficients: (a) Values (privacy-preserved vs. non-privacy-preserved algorithms); (b) Relative errors between privacy-preserved and non-privacy-preserved algorithms.
• Figure 4: Aggregate state and real indoor temperatures of zones.
• Figure 5: The first column of $\hat{\tau}_{in,z}$ and of sites encryption version $\hat{\tau}_{in,z} W^T$: (a) Iteration 1; (b) Iteration 2; (c) Iteration 3.

Theorems & Definitions (9)

• Remark 1
• Remark 2
• Remark 3
• Lemma 1
• Remark 4
• Proposition 1
• proof
• Proposition 2
• proof