Energy Case Study: Commercial Building HVAC System Optimization

Energy Case Study: Commercial Building HVAC System Optimization

Energy Case Study: Commercial Building HVAC System Optimization

According to the U.S. Energy Information Administration, HVAC systems comprise up to 56% of the overall consumption of energy in any commercial building. Therefore, achieving even small contributions to optimize the usage of those systems can provide impactful changes in the grid energy consumption as well as a building’s annual bill.

Artificial intelligence has come a long way in learning how to optimize systems, and this is something that is especially needed in the energy sector. More specifically, artificial intelligence has significant potential to reduce energy usage by analyzing a building’s temperature conditions, forecasting HVAC needs, and making recommendations for automated system controls that improve efficiency.

Perceiver AI had the opportunity to put this hypothesis to the test. A top New York based energy company contacted us for assistance in using artificial intelligence and data forecasting to optimize HVAC systems for commercial buildings. Their mission is to help clients achieve cost savings on their energy bill through intelligent demand forecasting as well as equipment changes on site. Here’s the process we used to arrive at an optimization solution for them.

The problem presented to Perceiver

The objective for Perceiver was to create a model in which a building’s HVAC system state could be off the most number of 15 minute intervals possible while still providing comfortable temperature conditions inside a room. The firm requested using setbacks, which enable the temperature in a room to shift lower or higher within a specified range in order to reduce HVAC system usage and thereby save utility costs. For the purposes of the test, the setback parameters were between 66 degrees and 75 degrees Fahrenheit.

They provided us with a three year HVAC dataset for a sample room. The data fields included outside temperature, room temperature, configured temperature by the user, occupancy metrics, and the HVAC system’s modes in which the room was configured for each 15 minute interval during the period. Perceiver’s job was to analyze the dataset and then develop a model that could be used to optimize the HVAC system serving that room.

Solving the problem

We decided that the problem needed to be divided into two pieces. First, Perceiver needed to be able to adequately predict room temperature given the current field conditions. Second, Perceiver needed to optimize the HVAC system to stay off for the most amount of time possible while achieving the desired temperature, as well as avoiding the setback points that incur energy loss.

For the first problem, we used one year’s worth of data in order to predict temperature conditions for that room for the next two years. That gave us an idea of how accurate Perceiver’s predictions were in comparison to the actual data. The chart below maps out the actual temperature over the three year span in gray and Perceiver’s predictions in orange. Our average error rate for the real vs. predicted temperature was only 0.5%. This means that with just one year of data, we were able to predict the room temperature based on current field conditions with a difference of just 0.5%.

After ensuring that Perceiver could correctly predict future temperature patterns given certain conditions, we tackled the second problem: transitioning the HVAC device to “off” the most number of 15 minute periods possible, without creating uncomfortable room conditions and while abiding by the setback points. In the chart below, we mapped out a one year comparison of the results between the real temperature based on the model of the data (the gray line) and the model of what Perceiver would have done if in use (the orange line).

As you can see, Perceiver would have been able to avoid high spikes of massive energy loss by adhering to the setback points within the 66 degree and 75 degree range.

The model was built looking for 95% accuracy to avoid longer processing times, so there is even more space for improvement. However, even with this margin of error, Perceiver determined that in just one year, the HVAC system within the room could have been switched “off” for a total of 1,838 15 minute segments. This translates to 459.5 hours, or roughly 20 days, of saved energy.

Extrapolating the solution

So what does this mean for a building’s energy use and its bottom line? Based on the findings from the sample room, we created an analysis of the potential savings of using Perceiver’s model to control the HVAC system, extrapolated to the whole building and then to the firm’s entire portfolio. We used the following assumptions in our calculations:

(a) Cost of energy per building per day = $0.21/kWh in NYC x ((816 kW x 12 hrs) + (160 kW x 12 hrs)) = $2,460
(b) Rooms per building = 100
(c) Cost of energy per room per hour = $2,460 / 100 rooms  / 24 hours = $1.03
(d) Buildings in the portfolio = 300

Combining these assumptions with the predictions from Perceiver, we estimated the following cost savings:

Savings per room per year (hours)

459.5 hours

Savings per room per year ($)

$473

Savings per building per year ($)

$47,300

Savings for all buildings per year ($)

$14,190,000

 

Perceiver AI showcases the power of artificial intelligence to analyze large-scale, complex datasets and generate solutions to some of businesses’ most pressing problems. Please contact us to discuss custom Perceiver AI solutions.