responsive to customer needs and changing environments. The ubiquitous broadband connection to the
Internet and the cloud allow embedded developers to realize these values. Using a hybrid model takes the best of both worlds by doing the heavy lifting offline, while maintaining critical pieces of functionality in the device. The technique is in finding the appropriate balance point in how much logic/memory lives where and how fast developers can adapt to evolving customer needs.

The anatomy of smart energy devices for residential applications begins with the tradeoff of brainpower versus cost. Consider the residential programmable thermostat, an embedded device that has been in production for nearly 50 years. Functionally very simple, the thermostat simply measures and maintains a household’s inside temperature. More sophisticated models include humidity sensors and possibly air quality sensors. The thermostat then controls the Heating, Venting, and Air Conditioning (HVAC) system according to the consumer’s temperature preferences established for specific times of the day. This is a coarse approach to alter temperature settings in order to save energy costs. Higher-end models have up to four set point changes that provide sufficient functionality for most consumer needs (serving the majority of abilities and attention spans).

Over the years, as HVAC systems became much more sophisticated with multi-staged compressors and burners as well as digitally controlled air handlers and blowers, additional control logic and signal processing was required in the thermostat, forcing the design to be upgraded from a simple microcontroller to a full-fledged microprocessor to handle the compute intensive processing requirements. Suddenly, memory management skills were needed with a larger DRAM and flash footprint. With the added requirement for Internet connectivity, a wireless subsystem with either Wi-Fi, ZigBee, or Z-Wave had to be designed in as well. And unless power can be drawn from the HVAC system, power management logic must be considered in order to meet the consumer need for at least one year without swapping out the batteries (Figure 1).

 

 

Smart thermostats in the Internet of Things age

As consumer requirements shift to home devices that are controllable by their smartphone and need to decrease energy costs in a more automated way, additional processing power is required to operate sophisticated and complex algorithms. However, should these real-time algorithms reside on the thermostat itself or rely instead on the data processing power available through cloud computing? As data scientists evolve their increasingly sophisticated models requiring algorithmic updates, can the embedded device handle the revised real-time processing needs?

Thermostats are entering the Internet of Things (IoT) adoption phase where consumers demand additional sophistication such as learning their daily temperature habits and adapting to their needs; energy efficiency algorithms must also be incorporated. However, the compute power and real-time requirements for these algorithms to produce sufficient gains incur significant additional cost to each unit. Further, the dynamic needs for each customer requires real-time updates. The horsepower and cost to execute significant savings causes designers to use expensive, multi-core processors to achieve the required gains. Alternatively, it may be more prudent to perform the processing in the cloud where Billion Floating Point Operations Per Second (GFLOPS) of compute power reside at a fraction of the cost, saving as much as $10 in the Bill Of Materials (BOM).

Handling the required data points while saving energy

To yield significant energy efficiency savings, several data points and controls contained within the thermostat are required as I/O to an energy efficiency algorithm. Programmable communicating thermostats store the following variables:

• Cooling setpoint
• Heating setpoint
• HVAC mode
• HVAC state
• Fan mode
• Fan state
• Hold state
• Manual override
• Relay status

This data set is the minimum required to comply with the ZigBee HA1.2 specification, or Home Automation version 1.2. However, regardless of the wireless protocol, these variables establish a set of minimum data points that any high yielding efficiency algorithm requires from the thermostat.

External data such as local temperature and humidity are also required as they affect the inside temperature and cause state changes in the HVAC system. For instance, during hot summers and cold winters, outside temperatures fluctuate drastically during the day, greatly affecting the ability of the home to cool down and heat up. Therefore, these variables will change frequently. Building a sophisticated model of the home’s thermodynamic properties requires storage of all of these variables and then processing them in real-time to provide sufficient resolution for the algorithm to yield valuable results. Rather than caching all of this time-series HVAC and external temperature data, it is more efficient to transmit and store this data in the cloud, and execute the most applicable algorithm there without burdening the local processor, or increasing the memory footprint to accommodate these data requirements.

Drawing from the data collected from millions of thermostats over many days and seasons, energy efficiency models are better equipped to predict individual household physical characteristics and reaction to consumer behavior under given temperature conditions and comfort levels as set by consumers. Algorithms and models will frequently be updated to incorporate this real-time feedback. It is far easier to change these algorithms in servers that reside in the cloud rather than transmit and store new algorithms in each thermostat.

Optimization with a hybrid strategy

With wireless connectivity fast becoming vital for product acceptance, many wireless subsystems have an embedded CPU that has sufficient Million Instructions Per Second (MIPS) to process wireless protocols while simultaneously possessing the headroom to operate the required thermostat logic. This removes the need for a second microprocessor, saving money and board space. However, while lower overall BOM costs are achieved, this tradeoff creates a lack of compute power for the energy efficiency algorithms. The real-time processing needs for the normal thermostat operation, for wireless networking, and for the efficiency algorithms, supersedes consumer-grade processors. Therefore, the right balance point is to design a very-low-cost connected and programmable thermostat that networks to a cloud-based service that optimizes the energy efficiency for the home and the individual temperature preferences required by its household members.

Ted Fagenson is Chief Marketing Officer for EcoFactor.

EcoFactor www.ecofactor.com

Follow: Twitter Facebook LinkedIn YouTube