Smart Comfort & Control

It is easy, deep in protocols and dashboards, to forget what all of it is for. A building conditions air, moves light and dampens sound so that the humans inside can work, rest and think without fighting their environment. Comfort is the served decision behind most of the smart-building stack - and it is worth stating plainly, because a twin that optimises energy while making people miserable has failed at its real job.

Thermal comfort is the best-studied piece. The classic engineering model is PMV/PPD (Predicted Mean Vote and Predicted Percentage Dissatisfied), which combines six factors - air temperature, radiant temperature, air speed, humidity, clothing and metabolic rate - into a prediction of how a typical group will feel. It is genuinely useful, but notice its own honesty: even at the theoretical optimum, PPD never drops to zero. A well-designed space still leaves roughly five percent of people dissatisfied by definition. There is no setpoint that pleases everyone, and pretending otherwise is the first mistake of naive comfort control. Comfort is not only thermal, either - light levels and glare, background noise and speech privacy, air freshness and even a view of the outside all feed the felt experience. The smart building touches all of them, which is exactly why comfort is the natural meeting point of every layer below.

There is a second reason to keep comfort front of mind: it is where the business case and the human case finally converge. Discomfort is not just unpleasant, it is expensive - it drives complaints that consume facilities time, it pushes people to bring in space heaters and desk fans that wreck the energy model, and in an office it quietly erodes the concentration the whole building exists to house. When a twin optimises purely for energy or equipment life and treats the occupants as a constraint to be tolerated, it tends to produce exactly these hidden costs. The reframe that runs through this entire module is to put the person back at the centre: energy, maintenance and carbon all matter enormously, but they are in service of a building that people can actually work, rest and think in. Get comfort right and the other objectives get easier, because you stop fighting the occupants; get it wrong and they will route around your clever controls one space heater at a time.

If one setpoint cannot please a whole floor, the elegant answer is to stop trying. Personal comfort systems (PCS) condition the individual rather than the entire zone: a small task fan for local air movement, a gently heated or cooled chair, a footwarmer, a desk radiant panel. Because they act within centimetres of the body, they need only a few watts to shift someone across a couple of degrees of perceived temperature - a fraction of the energy it takes to move a whole room.

The design move is powerful. You let the central system hold a wider, more relaxed band - say a cooler ambient in winter and a warmer one in summer - and let each person close the last gap themselves. Research on PCS consistently shows both higher satisfaction and lower energy, because the building stops overcooling in July to satisfy the coldest complainer and stops overheating in January for the warmest. In the smart-building version, those personal devices report back: the twin can see who is running their chair on full heat and infer that the ambient band is drifting too cold for that neighbourhood of desks. The occupant gets agency; the building gets data about real preference instead of guesses.

There is a well-known trap in comfort research: the placebo thermostat. Give people a dial that is not actually connected to anything, and complaints still fall, because the sense of control matters almost as much as the control itself. It is tempting to exploit that. Do not. A building that fakes agency erodes trust the moment someone notices, and someone always notices.

Honest smart comfort gives occupants a genuine channel and is transparent about what it does. In practice a tap on an app is rarely a direct hand on a valve - it is a request the system weighs against neighbours, safety limits and energy. That is fine, and better stated openly: warmer here nudges the local setpoint a degree for a while, then decays back. The system can also close the loop by telling people what happened - acknowledging the request, showing the current temperature, explaining that it is easing the zone up. Crucially, every one of those requests is a labelled datapoint. Thousands of them become a live comfort map: which corners run cold, which meeting rooms bake in afternoon sun, which complaints cluster near a failing VAV box. That is the twin earning its keep - turning subjective grumbling into a signal the facilities team can act on. Control given to occupants is not a loss of engineering authority; it is the richest comfort sensor you will ever install.

The PMV model assumes a tightly conditioned box. The adaptive comfort model - embodied in standards like ASHRAE 55 and EN 16798 - starts from a different, well-evidenced observation: people who can open a window, change their clothes and expect some seasonal variation are comfortable across a much wider range, and their comfort temperature drifts up and down with the outdoor running-mean temperature. In a naturally ventilated or mixed-mode building, chasing a fixed 22 degrees all year is not just wasteful, it is actually less comfortable than letting the interior breathe with the season.

This is where the honest tension lives. Individual control and building efficiency mostly align, but not always. Personal comfort systems and adaptive bands usually save energy and raise satisfaction together - the happy case. But sometimes they collide: a heatwave and a grid demand-response event might ask the building to let temperatures drift up precisely when the coldest-blooded occupant wants more cooling. A twin can broker this intelligently - pre-cooling thermal mass overnight, staging setbacks so no single space swings too far, and using PCS to keep individuals comfortable while the ambient relaxes - but it cannot repeal physics. The mature position is to state the trade-off plainly, give people the widest honest band plus local control, and reserve aggressive efficiency moves for moments that genuinely matter. Comfort optimised behind people's backs breeds mistrust; comfort negotiated with them, and evidenced by the twin, is what smart control is actually for.

All of this becomes far more powerful once the twin remembers. A single comfort request is a moment; thousands of them, logged with location, time and outdoor conditions, become a learned model of preference for a real population in a real building. This is where the analytics and control ideas from earlier modules meet the human layer. The twin can see that the north-east corner runs a degree cold every winter morning, that a particular meeting room overheats whenever the afternoon sun and a full table coincide, that satisfaction on one floor collapses above a certain humidity. None of that is visible from a design spec; it only emerges from operation.

With that history the building can shift from reactive to anticipatory comfort. Instead of waiting for the complaint, a model-predictive controller can use the weather forecast, the occupancy schedule and the learned preferences to pre-condition spaces just enough, just in time - warming that cold corner before people arrive, pre-cooling the sunny room before the meeting fills it. Done well, occupants simply stop needing to ask, which is the truest sign of good comfort: they are not thinking about the building at all. Two cautions keep this honest. First, learned preference is still a proxy - people change, seasons change, and the model needs to keep listening rather than freezing yesterday's guess into today's setpoint. Second, comfort data is personal data: knowing who requests heat, when and where, edges toward tracking individuals, so the same privacy discipline the rest of this module insists on applies here too. A comfort twin that learns respectfully is one of the most genuinely useful things in the whole stack; one that quietly profiles people is not.