Paul Gutekunst 0:07
Hello and welcome to utiliVisor's response to the NYSERDA PropTech Challenge. My name is Paul Gutekunst and I am a lead development engineer here at utiliVisor.
Paul Gutekunst 0:15
The challenge is about understanding how COVID-19 impacted building occupancy levels and tenant energy usage then to determine if there's any correlation between the two. Finally, we want to identify factors that go into accurately predicting tenant energy usage.
Paul Gutekunst 0:33
We broke down our approach down into four steps. First, we scrub and refine the data that is provided. Second, the data is used to develop predictive formulas and algorithms then analyze the results and adjust accordingly. Finally, our results are displayed on a platform that is easy to digest and actionable.
Paul Gutekunst 0:52
To take a deeper dive into how utiliVisor scrubs data, we import the raw data into utiliVisor's suite of analytics tools that will automatically normalize all timestamps and apply null field techniques. From there we can learn more about the data patterns and formulate assumptions on what this meter data is; using utiliVisor's data history and experience.
Paul Gutekunst 1:12
The scrub data is presented to the operations team where it can be visualized inside utiliVisor's platform again. The operations team will analyze the data and get a basic ideas such as occupancy levels, business hours, holiday schedule, and even when the cleaning crew arrives. Then the data is fed into statistical analysis tools that is partitioned by hour of day, day of week, month of year, and season, that will help identify outliers and visualize the data within the model.
Paul Gutekunst 1:39
The result is we can now confidently answer questions the NYSERDA PropTech Challenge presents.
Paul Gutekunst 1:46
You can visit our website to view our forecasts of consumption for all tenant usage meters on August 31, 2020.
Paul Gutekunst 1:54
So how correlated are building-wide occupancy and tenant consumption? From our analysis there is at least some. This correlation was dependent on tenant meters with notable increases in power during business hours.
Paul Gutekunst 2:08
We used R squared to measure the quality of our predictive model, we reached up to point seven nine r squared in our current prediction model.
Paul Gutekunst 2:17
So what predictors were most important in determining energy efficiency? From a tenant perspective, using time of day, day of week, building occupancy and outside air temperature, helped build our predictive models. From a building perspective, in early June 2020, as the building reach higher outside temperature, the ConEd Utilities steam consumers were shut down and had drastic impacts on building energy efficiency.
Paul Gutekunst 2:42
So which day has had the most energy efficient occupancy level? Pre COVID, this was on March 10, 2020. During COVID, the best day we had was on August 17, 2020.
Paul Gutekunst 2:53
What else can we conclude from our model? Well, we can see what the base load is for the building independent meters before and during COVID. We can see that the office cleaning crew was visiting the tenant space later in the day between nine and 11pm. Finally, we can get an idea of the tenant occupancy times based on the consumption data.
Paul Gutekunst 3:11
And last but not least, what would we need to better our current prediction model, the bigger items we would like is building and tenant square footage. Also, knowing the tenant type would have been useful whether it be an office building tenant, restaurant or server farm. And finally having tenant occupancy count by hour instead of building occupancy totals would help improve our models.
Paul Gutekunst 3:31
We enjoyed what we learned throughout this process. If you have any questions or comments, please feel free to contact us by visiting www.utilivisor.com.