Yes, you read the title correctly; it’s not a typo. Let me show you how we can control the room temperature in order to keep the chiller off. This is a follow up to my previous blog post “Simulation: A Safe Playground for Smart HVAC Control”
We continue to explore the idea of optimizing room temperature setpoints in order to 1) follow the unforced room temperature and 2) shift HVAC load to the “cheapest” period of a day (aka time of use or TOU). Within some comfort range, the optimisation finds the room temperature profile that requires the least HVAC energy (or cost) to deliver. We also continue to adopt the much simplified and high-level view of an HVAC system and assume complete knowledge on the amount of cooling/heating power required to maintain or change the room temperature under various internal and external conditions.
The room temperature is unforced when the chiller is off. The following animation shows two days of simulation (the x axis is measured in days). It shows that as we widen the comfort range, the optimized room temperature setpoint (zt_sp) becomes more and more like the unforced room temperature and, eventually, the cooling energy reduces to zero!
This is not at all unrealistic. During the shoulder months of the year, depending on the comfort range, the optimisation can reduce the number of days that the chiller has to be on; and if it has to be on, the optimisation can delay and/or minimize the hours that the chiller has to be on.
The following examples show how the optimized room temperature setpoint (zt_sp) shifts the HVAC load to time periods when TOU tariff is cheapest. The TOU tariff specifies how the energy price changes with hour-of-day. Note that the optimisation also takes into account how fast the room temperature changes with outside air temperature!
In order for this type of room temperature optimization to work, we need to make sure the underlying HVAC system performs reasonably well. Modern commercial building HVAC systems are much more complex than the simple model we assume in the above optimisation. There are many mechanical parts: pumps, pipes, valves, fans, ducts, and dampers. They aggregate into functional equipment: chillers, AHUs and VAV boxes. There are hundreds of control loops and a number of one-to-many supply-demand relationships of refrigerant, water, and air. In order to maintain the performance of the system, it’s important to:
No. 1 is no small task. To address it, we have scheduled on-site visual inspections and remote diagnostic systems that issue alarms. No. 2, however, is what an HVAC system is designed for, which is also good to be continuously monitored and validated. To further address No. 2, there are —for example— various SAT reset strategies.
We are essentially optimising the cooling demand when we optimise the room temperature. We rely on the fact that when cooling demand is reduced, cooling supply is also reduced. For example, when the room needs less cool supply air, the fan speed is reduced (save on fan power); more chilled water is bypassed; the chiller control logic sees less drop between supplied and returned water temperatures, and stages down or turns off the chiller when this drop is small enough. These are the basic functionalities of a variable air volume HVAC system. They are what the system is designed to do and they do not require the implementation of a different SAT reset strategy. However, following from the example, if the chiller was programmed to follow a pure time-based staging (i.e., on/off) schedule then it won’t be off even when the cooling demand is zero. Even in this non-ideal case scenario, we could still save on fan and chiller power.
For the purpose of this blog post, we used the TOU tariff as a signal to shift the cooling demand to the “cheapest” times of a day. In my next blog post, we will consider other signals such as chiller COP, economic cooling, and demand ceiling. Thinking back, it’s pretty amazing to see that all these different goals, including the ability to follow the unforced room temperature within comfort range, are achieved by mathematically optimising a single cost functional! This means the optimisation can handle a combination of these goals at once without the need to program one specific behaviour for each goal and each combination.
Dr. Rui (Ray) Xu is Data Scientist at BuildingIQ. During his career, he has encountered many challenges in the transfer of human knowledge into machines, the interpretation of results of algorithms into human intuition, and the verification of evolving strategies. This experience has helped him to solve some of the most difficult and interesting problems in the industry. He describes himself as a bit quiet but also very energetic and assertive when there is a mathematical, simulated and/or experimental proof behind the scenes.