Control of thermostatic loads using moving horizon estimation of individual load states

Recent research has shown that thermostatically controlled loads (TCLs) can provide power system services. However, a key challenge is to achieve coordinated control of large populations of resources using existing communication and control infrastructure or with minimal addition of new infrastructure. In this paper, we assume that we only have access to realistic measurements, i.e. data from residential smart meters every 15 minutes and noisy real-time measurements of the aggregate power consumption of TCLs from distribution substations. Our contribution is to develop a moving horizon state estimator (MHSE) to estimate the states of individual stochastic TCLs from these measurements. This is in contrast to previous work that focused on estimating the states of aggregate system models. The proposed MHSE is benchmarked against a simpler model-based prediction. We also propose a scalable closed-loop control structure that uses the MHSE method to provide frequency control with TCL populations. We demonstrate our results via a number of case studies with different TCL aggregations, process and measurement noise characteristics, and controller forcing levels. Our simulations show that the MHSE generally provides accurate state estimates and improves the controller performance.