AI-based Automation for Commercial Office HVAC
Persistent, automated, occupant-responsive HVAC optimization
- Fortune 500
- Project Title
- HVAC Optimization for SF Office of Fortune 500 Company
- October 24, 2018 00:00:01 - October 31, 2018 23:59:59
- Identified 19% energy savings while increasing productivity by $300k through HVAC optimization
Verdigris is an energy intelligence and building automation service.
Building automation has three requirements: accurate energy and occupancy forecasting, building context, and an operator objective.
Verdigris sensors monitor energy usage thousands of times per second. To create forecasts our AI adds local weather, utility pricing, building management system (BMS) data, and other available datasets. Where data may not be explicitly available, Verdigris AI can infer it. In this simulation, Verdigris’ AI learns the building’s occupancy by observing security badging patterns.
This report simulates the results of an automatic and persistently optimizing heating ventilation and air conditioning (HVAC) system for a Fortune 500 operated building. These control schemes may be tuned by facility operators for both occupant comfort and cost efficiency.
Because Verdigris measures a building’s energy usage in real-time, these efficiencies are quantifiable, auditable and trustable.
Building Industry Classification: Office / Laboratory mixed use
- Weather (real time)
- BMS (historical indoor temperature
- Energy - real time, 8khz per sensor
- Utility Pricing
- Indoor Temperature Forecast
- Energy Forecast
- Occupancy Forecast
- Energy - Persistent automated HVAC energy savings up to 18.7%
- Energy Cost - Persistent automated HVAC cost savings 22.7-33.7%, depending on optimization criteria.
- Comfort - Opportunity to increase from currently 4.5% of “occupied hours” within ASHRAE-55 operating standards to a persistent automated 100% of time within optimal productivity performance.
- Productivity Impact - Productivity value of at least $300k identified based on publicly available salary data and conservative occupancy estimation. Sample calculator attached.
- Project Payback - The combined value of energy savings and increased productivity will result in a project payback period of 1 year.
- Return on Investment - The overall 5 year ROI for the project will be 5x.
Today’s state-of-the-art building and environmental control systems have failed to enable facilities and operations management. Our buildings are inefficient and the people using them are underserved.
Building infrastructure and management technology has not kept pace with the the increasing complexity of the built environment. Building energy needs and occupant needs are increasingly dynamic. Contributing extrinsic factors include seasonal building usage, daylight savings, or weather phenomena. Building run longer hours, support wider end uses, and support greater levels of economic productivity, creating thinner margin for error.
Furthermore, split incentives create barriers to invest in solutions. For example, it is common in leased facilities for landlords to install the cheapest HVAC equipment available. They know it will be renters, not themselves, who pay utilities and maintenance. Inadequate equipment becomes the tenant’s problem. These conflicted incentives have significant downstream impacts on energy efficiency, carbon emissions and people.
Labor costs to support brute force corrective actions have been prohibitive. To meet occupant comfort and maintain cost and energy efficiency, a dynamic machine-assisted approach is needed.
Verdigris has developed a solution to automatically minimize HVAC energy usage while maintaining occupancy comfort, adaptive to both intrinsic and extrinsic context. Our model provides three (3) operator-tunable parameters. These parameters weigh how much the user operator cares about:
1. energy costs
2. stable operating temperature
3. occupant comfort
The solution is cloud based, enabling 24/7 continuous access for users to the simulation results and for environmental systems to the service.
Verdigris artificial intelligence engine builds the model based on real world data.
- Verdigris sensors provide real time energy.
- Verdigris uses utility pricing based on a custom rate structure.
- Verdigris uses reliable third party weather information provider.
- Verdigris uses Fortune 500 Company’s badging information to estimate building occupancy.
- Verdigris uses ASHRAE 55 compliance models and Fortune 500 Company’s BMS data to model building comfort requirements.
The Verdigris’ model provides the following key features:
- Users are able to receive a time-series occupancy estimate for a building.
- Users are able to receive an HVAC optimization plan based on occupancy and cost reduction targets.
- Users are able to receive advice on HVAC optimization compliance with ASHRAE thermal comfort guidelines.
- Users are able to receive monthly reports demonstrating the dollar impact of having implemented HVAC optimizations.
Example ASHRAE-55 Comfort Region in Blue
We ran our model over a single week to demonstrate daily variance4. The model was run with three parameter selections:
1. optimizing for temperature stability in an occupant comfort temperature zone
2. optimizing for occupant responsiveness in an occupant comfort temperature zone
3. optimizing to balance objectives
The following results show the energy consumption of the three different strategies compared to actual energy consumption measured by the Verdigris energy meters.
Selection 1: Optimizing for temperature stability in an occupant comfort temperature zone
The red charted line is the summation of all HVAC in the building and shows actual HVAC circuit energy consumption for the simulated period. The model result is shown in orange. The model shows the energy consumption when HVAC is optimized for maintaining a stable indoor temperature in an occupant comfort zone.
Selection 2: Optimizing for occupant responsiveness in an occupant comfort temperature zone
The red charted line is the summation of all HVAC in the building and shows actual HVAC circuit energy consumption for the simulated period. The model result is shown in green. The model shows the energy consumption when HVAC is optimized to be responsive to building occupancy.
Selection 3: Optimizing to balance objectives
The red charted line is the summation of all HVAC in the building and shows actual HVAC circuit energy consumption for the simulated period. The model result is shown in blue. The model shows the energy consumption when HVAC is optimized to balance6 the considerations of stable temperature and occupant responsiveness.
Thermal Comfort Performance As-Is
Based on ASHRAE 55, we analyze whether indoor temperatures with current HVAC operations will locate within the occupant comfort region. We selected a similar week April 5th to April 11th for examination7.
The following figure illustrates the comfort indicator during that period based on BMS indoor temperature. A value of 1 indicates the temperature is within comfort region. A value of 0 indicates the opposite. Analysis indicates that the facility operates only 10.5% time within comfort zone.
The following plot is a histogram of indoor temperatures during that time period. It shows an average around 21C (or about 70F). ASHRAE 55 specifies a lower bound at 23C (73.4F). The indoor temperature appears overcooled for thermal comfort, which may impact productivity.
Thermal Comfort Performance To-Be
Based on ASHRAE 55, we estimate whether indoor temperatures under Verdigris AI-optimized HVAC operations will locate within the occupant comfort region.
The following figure illustrates the comfort indicator for all three Verdigris optimized selections during the simulation period.
For both stable temperature and balanced objective scenarios, the AI engine strictly prioritizes time in ASHRAE 55 comfort regions, resulting in an estimated 100% compliance with comfort objectives. In an occupancy responsive optimization scenario, we estimate some time intervals during which the temperature may fall outside comfort zone guidelines. All time intervals spent outside comfort zones happen during times the AI engine forecasts little to no occupancy.
CurrentTemperature StabilityOccupant ResponsiveBalanced ObjectivesTime in ASHRAE 55 comfort region Total10.5%100%89.3%100%Time in ASHRAE 55 comfort region Occupied Hours84.5%100%100%100%Time in ASHRAE 55 comfort region Normalized to Occupancy3.2%100%98.8%100%“Uncomfortable” hoursNot analyzednone1am-7am, Oct 27th, Saturday; 1am-9am, Oct 28th, Sunday; 4am Oct 29th, Monday; 4am Oct 31th, Wednesday9none
All models indicate the potential to optimize the current HVAC systems at Fortune 500 Company building. For example, HVAC systems continue to operate at levels more than necessary to maintain a comfortable occupant environment during the weekends when occupancy is minimal. By tuning parameter weights, Verdigris’ model can provide different suggestions for different building needs.
HVAC Cost and Savings Tables Total heating, ventilation and air conditioning costs for the demonstrated week for each scenario are summarized. Scenarios incorporate customer electricity rate structures.
ActualTemperature StabilityOccupant ResponsiveBalanced ObjectiveskWh2087217416951702Cost$424.03$327.89$281.03$281.98
Net savings compared with actual usage are summarized. Comfort values are negative because this strategy uses more energy to stabilize an indoor environment within the comfortable temperature zone. All strategies show net dollar savings due to peak demand reduction.
ActualTemperature StabilityOccupant ResponsiveBalanced ObjectiveskWh0-87.5391385CO20-54243238CH40-3.214.414.1N2O0-0.391.761.73CO2e0-54.4243240Fossil CO20-82.7370364Energy Savings0-0.04%18.7%18.4%Dollar Savings0$96.14$143.00$142.05Dollar Savings (%)10022.7%33.7%33.5%
Forecasting internal building temperature and normalizing for weather
An effective building automation solution must forecast indoor temperatures to manage the impact of external weather conditions. Modern BMS systems can provide historical data but do not provide indoor temperature forecasts.
Verdigris uses statistical methods to map the impact of outdoor weather on indoor temperature. Indoor building temperature can then be inferred from weather forecasts. The following figure shows the scaled difference between indoor temperature forecast and the most comfort temperature for the demo week.
This pattern demonstrates the periodic variance between weather and indoor temperature. It matches intuition. The delta between indoor and outdoor temperature for Fortune 500 building is minimal during the daytime and widest at night time.
Occupancy is derived using a stochastic model. The model estimates how long people will stay in the building using a Gamma Distribution. The Fortune 500 building badge-in data is combined to generate an occupancy estimation.
In the plot below, weekend days with lower peak occupancy are visually identifiable. The shape of the occupancy curves appear roughly similar. People appear to badge into the office at similar times but in fewer number.
Impact Energy and Utilities Facilities and Equipment People Smart Buildings / Corporate alignment (Marketing and Strategy) Environment Value of continuous data access The model can better simulate real world conditions by acquiring BMS data for time periods that overlap with Verdigris energy data. This enables the model to better understand how energy use impacts building comfort metrics and enables more accurate HVAC modeling. In this model, HVAC operating modes are inferred by power usage. Integrated BMS and HVAC data enables modeling from actual HVAC operations. These results can be embedded in model for more precise suggestions. Future services Automation models may be paired with [Verdigris Intelligent Alerts] to automatically monitor for anomalies outside set operating parameters. Buildings with compatible building management systems can implement Verdigris intelligent control algorithms to automate occupancy responsiveness, temperature stability and cost efficiency. Control algorithms and optimization models can be developed for additional facilities or incorporate additional criteria.
1. A stable operating temperature can sometimes be an important process requirement. For example, in a data center or factory.
2. Both energy usage cost and peak demand cost are considered. Rate schedule is confidential and not published.
3. Historical indoor temperature is captured by BMS data export and can be automated through integration. Future indoor temperature is forecast using methods outlined in appendix.
4. With more occupancy and BMS data weekly, monthly or seasonal results could be shown. Occupancy data was modeled from a previous year. BMS data from April.
5. A scenario optimizing only for energy efficiency is intentionally omitted. This scenario would show that “everything off” would be the optimal method to save energy.
6. Balanced scenarios should be the most frequent case in practice. For example, some level of care is given to occupants in late evenings or weekends.
7. Fortune 500 Building BMS data set is constrained to end of April 2018
8. 9am - 7pm MTWTHF
9. All “uncomfortable” time intervals in occupant responsive scenario are located in either midnight or weekend hours when the occupancy is low.
10. % $ is calculated as savings $ divided by actual cost
11. 1 represents the inferred historical maximum occupancy during the period of available badging data. 0 represents the minimum.