Many of us are happily using advanced measurement and verification (M&V) methods, a.k.a. normalized metered energy consumption (NMEC), to quantify savings in our energy efficiency projects, now that various meter-based incentive program offerings are underway. My previous posts outlined why this is a ‘happy’ place for many commercial, institutional, and educational buildings:
- Energy savings are quantified in a way a customer understands them: at the meter.
- The methodology’s use as a savings tracking and persistence tool
- Lowered costs due to standardized well-known methodology without perpetually updating requirements.
- Deep savings
- Scalability of approach
To be sure, there are significant issues to address as we do more projects in different buildings and facilities, including:
- assuring buildings are a good fit for an NMEC approach (predictable energy use and ‘discernable’ savings),
- customer acceptance of incentives based on performance, and
- accounting for the impact on energy use and savings of building changes unrelated to the efficiency measures (referred to as non-routine events or NREs).
Today we will discuss the current COVID situation, or maybe COVID era is a better word as its effects look like they will continue for an extended period. Businesses have shut down and their operations have been reduced dramatically as people stay at home to shelter in place. For example, energy use patterns for commercial buildings post-COVID look like patterns we normally only see on weekends. We seem to be through the worst of it and now businesses are gradually re-opening. Whether they will get back to their original capacities remains unknown.
Using meter-based approaches we make empirical models of energy use, usually statistical regressions, using data from drivers of energy use. In commercial buildings, the dominant variable is the ambient temperature for which the relevant data can be collected from on-line sources. Often temperature is enough, but in many cases other important drivers of energy use are needed such as daily building operation schedules, occupancy loads, operation modes, and production rates.
Data for these energy use drivers must come from the building. For example, daily operating schedules may be obtained from the building operator or inferred from the data. Building operation modes, such as a vacation or summer session in a k-12 school are readily available from the school’s posted calendar. Production rate data is rarely needed in the commercial sector but may be available in some form from the few facilities where production is relevant, such as restaurants.
Data on building occupancy is rarely available at the frequency used in an advanced M&V model (such as hourly or daily). Also, if the driver of energy use does not change substantially over the period an energy model was developed, it does not help ‘explain’ the energy use. In other words, it doesn’t have a valid statistical “link” with energy use.
For site-level NMEC, the analysis is customized to a particular site and its drivers of energy use. The major drivers must be identified and, to the extent that data can be collected, they should be included in the energy model development. This means we test that the model goodness of fit is within accepted criteria, assure the model coefficients are statistically valid, and assess whether the model is correctly specified. The approach helps us to understand the level of influence that each energy driver has and helps reduce the level of uncertainty in the final savings estimation.
As mentioned above, except for weather data, the data source for each potential driver is the building itself. Often it is not possible to collect continuously varying data at the same time interval used to develop site-level baseline models (such as sub-hourly, hourly, or daily). Even if it is possible, the values must vary widely over their possible range in order to test whether they indeed influence energy use.
In cases where continuously varying data is not available, indicator variables may prove sufficient. Indicator variables take values of 0 or 1 over the model development period to indicate the presence or absence of a condition. Taken alone they represent a constant shift in energy use, however they may be combined with other independent variables such as temperature if the condition has temperature dependence. Operation modes, occupancy levels, and NREs can be modeled successfully with indicator variables. Indicator variables are not required to have the same time granularity as the energy and temperature variables.
Thoughts on COVID and Advanced M&V
This background helps us begin to understand how to address the impacts of COVID in our commercial building projects. COVID has lowered commercial building energy use over these past months because people are staying home and social distancing. Building operators have dialed back building lighting and HVAC services to avoid unnecessary energy waste.
For site-level NMEC projects that are in their performance phase, this is a real problem. If our baseline models were developed over a wide range of occupancy load conditions (as well as under a wide range of weather and other conditions), our models could accurately determine ‘what baseline use would have been’ under the present COVID conditions and reliably determine savings (though we might expect the savings to be small in the COVID period). But we typically do not collect occupancy data because it did not vary significantly in the baseline period. Other approaches must be taken to account for the low-occupancy/low-energy-use performance period so that savings can be accurately determined. These approaches may include extending the performance period, or developing a model from performance period data and using baseline period conditions to determine what performance period energy use would have been under baseline conditions (a process known as ‘backcasting’). Additional ideas are being tested by industry experts.
However, the situation is different for new NMEC projects. We can develop energy models now and use occupancy load data – which varies considerably over the year including the COVID period – and develop accurate baseline models. Because these models have a wide range of not only temperature and occupancy load data, they will be accurate when used to predict ‘what baseline use would have been’ in future periods after measures are installed and as we return more or less to business as usual. Another approach might use indicator variables for COVID-period operating modes, which may create statistically significant parameters in the baseline model.
In summary, we may be able to use the same techniques that underlie the NMEC approach as a means of describing building energy use under unusual conditions, like the COVID shutdowns.
If you’re concerned about site-level impacts in the post-COVID era, let us know. We’ll be following progress carefully and continuing to evolve our M&V methods to adapt to changing baseline conditions.