Weather Regression

There is insufficent data to do regression on weather at the monthly HDD/CDD level. A possible analysis would use HDD/CDD on daily or hourly kWh and even to filter the data for occupancy, but this analysis was out of scope for now. We did take a quick approach to show some interesting correlations. Expand the "Show Data Table" section above, and note that the rightmost two columns show R² for monthly kWh on HDD and CDD. In this quick approach, we make up for the lack of data by defining the heating periods as three winter months (Dec, Jan, Feb) plus all six shoulder months (Mar, Apr, May, Sep, Oct and Nov). Similarly, for CDD we use all shoulder months plus the three cooling season months (Jun, Jul, Aug). We took this approach because the shoulder months in many cases show significant HDD and/or CDD as shown in the first two rows of the table.

Using the data table above, we note high correlations (highlighted in red) on certain circuits, which in most cases are fairly obvious:

• Lighting Panel 31 correlates very closely with HDD. While we normally don't expect lighting to correlate with weather, the shorter daylight hours during the winter months can have a "weather" effect on lighting. Without specific knowledge of the equipment on LP 31, we cannot conclude anything except that there is an apparent correlation.
• Pump #1 correlates well with HDD, and Pump #2 correlates fairly well with CDD. This is proabably obvious if Pump #1 is related to a heating loop and Pump #2 is related to cooling.
• Smaller circuits also show correlation to HDD, such as Men's Room Heat, Back Stairwell and the Mechanical Room. It's worth casually noting the apparent correlation of the Elevator to CDD. Without knowing the demographics of occupancy in the cooling season, it would be a stretch to conclude anything but it does seem like people might be more likely to use the elevator during warm weather.
• The notable expected correlations are the Cooling Tower and RTU with CDD. We will discuss this further below, because the coincident operation of these two assets drive the peak demands during the summer.

The above statements may be far more conclusive than the data permits, but we do think that several weather-dependent relationships certainly exist.

Peak Demand

Discussion

In the visualization above, we take the hourly kWh values and sort them ascending from low to high (left to right). The chart can be dynamically filtered by clicking on the circles in the legend. When the chart is first loaded, we see that the base load for the Library is approximately 4.8 kW, signified by the left edge of the data where all values are at or above 4.8. In other words, when the Library is in its most shutdown state, it still demands 4.8 kW.

At the right edge of the data, we see that in 5% of the hours during the period, the demand was 28.2 kW or higher. The area under that data amounts to 10.3% of the total energy; when the Library is running at full tilt, it consumes about twice the energy of its average. In terms of peak demand mitigation, the hours at the most extreme right edge of the chart are of most interest. Those are the hours that the utility will use to calculate the peak demand charge on each month's bill.

We know that Gleason for all intents operates on a six-day schedule. If we click on "wd", all weekday lines are removed, leaving only weekend data. We see that the curve holds it's approximate shape, but the average of the data drops to 12.3 kWh (per hour). When we click on "sat" to remove the Saturday data points, we see that the shape changes dramatically - only Sunday data remains, and we see the average drops dramatically to just 10.5 kWh/hr.

A better version of this visualization would allow filtering for months, seasons and a true occupancy schedule. Over time, as AEI collects more long-term data, another enhancement could be year/year and season/season comparisons.

Monthly Peak Hours Identified

Using our analysis of the SiteSage data, after correcting for a flaw in their API that calculates kW, we have identified the peak demand period in each of the months where we have data. We use these peak hours to make a proxy calculation of the Gleason electric utility bill, and do a quick assessment of the effect of demand charge increases forecast for 2016. Finally, we show the underlying components of the equipment in use that is driving the peak usage, and with this we may be able to schedule equipment in a way that minimizes the peak demands.

We used the rates in effect on the December 2014 NSTAR bill for the Library.

The two rightmost columns show the potential impact of a doubling or quadrupling of the utility demand charge. The demand charge averages 60% or more of the average monthly bill, so these charges are significant. In terms of reducing electric energy usage, the focus should be on reducing the peak demands and less so on the average usage.

Click on "show all details above" to see the underlying circuits that most contributed to the peak demands in each period. The tendency is that during winter months, lighting dominates the peak demands. In the cooling season, the peaks are dominated by the RTU and Cooling tower. Lighting Panel 22 is a consistent Top 5 contributor to peak demands. In terms of demand response, we should learn more about the load on LP22 as a potential candidate for reducing the peak demand, as well as lowering the overall average load at the Library. During summer, if the RTU can be decoupled from the Cooling Tower, peak mitigation might also be achievable during those months.

Twitter