As students of UC Davis, we wish and hope for a sustainable future for the wellbeing of our campus and alumni. Achieving this goal requires the consideration of sustainable development while being environmentally conscious. We chose to analyze energy usage relative to campus infrastructure in anticipation to future energy demand needs.
What it does
Based on the csv, with specefic columns specified ( they're really just the default csvs of CEED with no modification) the user inputs, the dashboard provides a time series plot of time versus total usage per building class label on tab #1. Tab #2 provides what we feel is is lacking in the correct display, a forecast with "months ahead," a parameter that can be specified is located on the left sidebar if the user is interested in interpreting the behavior of usage any duration of time ahead.
How I built it
We used python and Jupiter notebooks to extract time series data for different buildings around campus through OSISoft's API. This data was then used to train sARIMA forecast models in R. Exploratory analysis revealed inconsistencies in the existing fluctuations of the data. We performed differencing operations to remove trend and account any seasonality components. Optimizing the ARIMA model parameters based on AIC/BIC criteria, we were able to forecast the user specified "months ahead" values. To present our findings, we gathered our results in a shiny dashboard that displayed initial time series plots and their forecasts. Instructions for navigating the interactive dashboard are detailed on the left sidebar.
Challenges I ran into
The unstable internet connection made it difficult to extract data using Jupiter notebooks. Since the data streams would be interrupted and the process had to be started again. Another issue was formatting an appropriate time series object within R.
Accomplishments that I'm proud of
Collectively, we're proud of our collaboration as a team. For many of us, this was our first Hackathon. Working efficiently required communication through Github. Conceptually, it was a bit of a challenge coming up with a workable and deliverable tool that would make sense to a user. We are especially prideful of our first interactive dashboard, a concept we have not yet seen in class.
What I learned
The quantities for which energy is consumed per building or per type for any given year. The cost required to maintain such resources can be optimized. We have the energy data as well as the building's metadata to achieve the answers we wish to better explain through time series forecasting. Insights gained from this will allow us to predict future demand and ideally, supply the optimal amount to conserve energy and reduce waste.
What's next for Forecasting UCD Buildings Energy Usage
Further exploration, we would like to experiment the cross correlation function with buildings against any influential factors that may affect how cost is measured and resources are allocated. For instance, weather, as in important seasonal aspect in time series, can lead energy consumption. We are also aware that electricity is independent, as noted, and would intend to discover if there is a mismatch between this particular energy and steam/water. It is our understanding that demand versus usage is a cause for inefficiency, with summed differences >0 being labeled as successive supply for the current demand, therefore inefficient.