The system comprises of a large number of Thermocloud clients that collaborate with each other through a Thermocloud server. Each client has its own unique thermal profile and operational characteristics of its HVAC systems. This includes a local algorithm, which runs on each individual thermostat, and a collaborative algorithm, which runs on a server to determine most relevant entry under current operating conditions.

Local algorithms are operated on each individual thermostat. It uses the user-defined schedule to learn optimal temperature range, and optimizes thermostat operation for that range. Moreover, it maintains a thermal profile that is unique to each individual house, and optimizes performance for this thermal profile. There are three main operations – probe, transit, and sustain.

Collaboration is done using a two-tiered approach. The first tier consists the cloud server, which distributes the problem space of entries into various servers. This is done using a variation of a K-nearest-neighbor algorithm with weighted Euclidean distance, that uses a server map to define the boundaries of each entry. There are also various expansion, splitting, and re-learning algorithms to modify this server map.

The second tier consists the individual servers. Once the first tier routes each entry to the designated server, this server determines the best collaborative machine learning algorithms(grad-learn, neural, or hybrid) to determine the most relevant entry for re-learning.

There are a tremendous number of combination of factors that affect optimal HVAC performance. Some of these factors are based on the user’s schedule, and operating conditions, while others are based on user zone configuration, as represented by the visual below.

These factors can add up to excess of 2.3 x 10³⁴ combination of factors for each house. As evident by these tremendous combinations, no two houses can operate in an identical manner. In order to optimize HVAC performance, it is crucial to optimize for these unique combination of factors. Identifying each house’s unique combination of factors is only possible using machine learning.

However, it is not computationally feasible for a small, isolated thermostat to adapt to this large combination of factors. Doing so would take tremendous time, computing power, and resources, and hence, would not offer savings from the very first day. Moreever, it is very likely that another similar house zone elsewhere had already learned and adapted to these optimal conditions using local algorithms. If this other thermostat shares it’s optimal conditions with the current thermostat, it can guide this thermostat to determine it’s optimal performance. Such collaboration between thermostats is the only way to save energy from the very first day.

Effective Cost Units are a mathematical model that normalizes the cost of running an HVAC stage for continuously running HVAC stage 1. An effective simulation requires using correct HVAC settings. Higher stages use greater energy than lower stages when running for the same amount of time. Having more startup-shutdown operations also increases energy usage. This conditions can be simulated with a stage rate and cycle cost.

\[ E_c(secs) = (c \times n) + \sum_{i=1}^n (w_{si} \times d_i) \]

where, \( E_c(secs) \): Effective Cost Units in seconds, \( c \): effective cost for a startup/shutdown cycle, \( n \): number of startup/shutdown cycles, \( i \): a startup/shutdown cycle, \( s \): HVAC stage that was used to run during a cycle, \( d_i \): Time duration when HVAC stage ran during cycle i, \( w_{si} \): Unique weightage for HVAC stage s during cycle i - with typical values being \(w_{1i}\) = 1 when stage s = 1, \( w_{2i} \) = 1.3 when stage s = 2, and \( w_{3i} \) = 1.6 when stage s = 3.

Thus, effective cost units provides a unique mathematical basis for comparing the effective cost of running an HVAC system.

A Thermocloud system running in a target zone identifies the operational environment of the HVAC system using an Active ECU. Active ECU is the effective cost unit for the target zone to change its temperature by one degree under a specific operating condition using the stage 1 of its HVAC system.

This enables representing the HVAC system’s capability to operate in a target zone and facilitates correlating its performance with similar target zones. Its value depends on the HVAC system, thermal profile of the zone and current operating conditions.

A Thermocloud system running in a target zone identifies the thermal environment of the target zone when HVAC system is not running using a Passive ECU. Passive ECU is the duration in which the target zone changes its temperature by one degree when the HVAC system is not active.

This enables representing the rate of loss of energy for a target zone and facilitates correlating its performance with similar target zones. Its value depends on the thermal profile of the zone and current operating conditions.