A Hardware-in-the-Loop Experimental Testbed using Air Conditioners for Grid Balancing

Abstract

Driven by the need to offset the variability of renewable generation on the grid, development of load control is a highly active field of research. However, practical use of residential loads for grid balancing remains rare, in part due to the cost of communicating with large numbers of small loads and also the limited experimentation done so far to demonstrate reliable operation. To establish a basis for the safe and reliable use of fleets of compressor loads as distributed energy resources, we constructed an experimental testbed in a laboratory, so that load coordination schemes could be tested at extreme conditions. This experimental testbed was used to tune a simulation testbed to which it was then linked, thereby augmenting the effective size of the fleet. Modeling of the system was done both to demonstrate the experimental testbed's behavior and also to understand how to tune the behavior of each load. Implementing this testbed has enabled rapid turnaround of experiments on various load control algorithms, and year-round testing without the constraints and limitations arising in seasonal field tests with real houses. Experimental results show the practical feasibility of an ensemble of small loads contributing to grid balancing.

Figures (12)

• Figure 1: Hardware-in-the-Loop experiment architecture.
• Figure 2: Left: Model houses are constructed inside a four foot cubic foam box. The internal heat source consists of a hydronic loop with an electric water heater. A duct fan forces air through the heat exchanger and mixes the room air in addition to the AC's fan. Right: The 20 units are stacked in pallet racks, minimizing cable lengths and overall footprint.
• Figure 3: (a) Inrush current measured for a single AC. (b) Time evolution of active power for a single AC. (c) Peak power consumption of an AC varies with ambient temperature. (d) The line voltage is uncorrelated with ambient temperature in the lab. (e) Histogram of temperatures for all 20 model houses, measured every second and integrated over several hours, while ACs have their compressors on and (f) off.
• Figure 4: The new ETP model of a model house. The basic ETP model from etp-model is on the left. In this basic model, $\dot{Q}_a$ would represent the AC as simply a step function with heating rate $\dot{Q}_a = 0$ when the AC is off and $\dot{Q}_a = -\dot{Q}_c$ when the AC is on. Here, though, we extended the model to explicitly include a simple model of the window AC to account for the time-varying power draw in cooling the house. Including this lossy Carnot refrigerator model introduces physics missing from the basic model, such as the time lag for the AC's heat exchangers to warm or cool and the effect of the outside temperature on the active power consumption of the AC (cf. Fig. \ref{['fig:testbed-combined']}(c)).
• Figure 5: (a) Cycle duration vs. internal heating rate for the experimental ACs (lines) and for the extended ETP model (circles). (b) Cycle duration vs. thermometer placement for the extended ETP model (line). As $1-f_{Hm}\rightarrow 1$, the thermostat uses the mixed air temperature in the middle of the model house, and the air temperature is weakly coupled to the water temperature so that the AC cycle durations are short. As $1-f_{Hm}\rightarrow0$, the air temperature is tightly coupled to the water temperature, and the entire heat capacity of the water must heat to $T_+$ and cool to $T_-$ in each cycle. Circles are data from an experimental AC run with its thermostat sensor at four different positions above the air-water heat exchanger.