A battery model can produce an impressive graph from a flawed load file. Missing intervals, mismatched time zones and confused power-versus-energy units may change the apparent peak pattern before a single equipment parameter is entered. Interval load data cleaning is therefore part of procurement due diligence, not merely spreadsheet housekeeping.
The objective is not to make the graph smoother. It is to preserve a defensible record of what the meter measured, what changed during preparation, and which assumptions remain. A clean-looking profile that hides a production peak can be less useful than an untidy file with honest quality flags.
Keep the raw evidence and its meaning together
Retain the original export unchanged. Record the meter location, measured quantity, units, interval duration, timestamp convention, time zone and whether the readings represent import, export or a net value. Include any known meter replacement, CT change or production shutdown that affects interpretation.
The Green Button Connect My Data framework supports authorised sharing of interval usage and billing data where implemented by the utility. It is an access framework, not a reason to assume every exported profile has the resolution or scope needed for a battery study.
Before applying the peak-shaving battery sizing method, establish whether a value is average kW during an interval, energy consumed in that interval, or a cumulative energy register. These quantities need different preparation. A column heading without its export documentation may be ambiguous.
Prepare a fault register before repairing anything
| Issue | How it can distort a battery study | Action to document |
|---|---|---|
| Missing interval | Hides demand or invents a recharge opportunity | Flag the gap and retain the chosen treatment |
| Duplicate timestamp | Counts an interval twice or shifts subsequent records | Identify source duplication versus a clock convention |
| Time-zone mismatch | Moves demand relative to PV or tariff periods | Convert consistently and preserve the original offset |
| Cumulative meter reset | Creates an artificial negative interval | Reconcile replacement or rollover evidence |
| Unit or scaling change | Creates a false step in power or energy | Verify export units and meter configuration history |
| Unexplained high reading | May be a real costly event or a measurement fault | Investigate before removal |
A high value is not automatically an outlier to delete. Check production records, alarms and meter history. For demand-charge analysis, the unusual interval may be precisely the one that sets the bill. Removing it without evidence can make the proposed savings easier to achieve only on paper.
DOE’s EMIS capabilities guide discusses interval-meter visualisation and gap-backfilling functions. For a purchase study, record which values were backfilled and why; do not present software-reconstructed readings as direct measurements.
Match the file to the model’s calendar conventions
Weather, load and tariff schedules must refer to compatible time conventions. A file recorded in local civil time may contain daylight-saving changes, while a modelling tool may expect a standardised annual sequence. Converting one series and leaving the others unchanged can shift the assumed solar surplus or charge window.
SAM’s electric-load documentation specifies average-kW time-series input and the row counts below. It also describes calendar assumptions that do not include daylight-saving time or leap years. Confirm the documentation for the tool and version used in the delivered study.
| Resolution | Load-value rows | Total file rows including header | Quantity represented |
|---|---|---|---|
| Hourly | 8,760 | 8,761 | Average kW for each time step |
| 15-minute | 35,040 | 35,041 | Average kW for each time step |
These counts describe the documented import format, not a universal rule for every meter file or simulation package. Do not delete a leap day, repeated clock hour or unmatched record merely to reach a target row count. First decide how the selected model represents time, then record a justified transformation.
Interval-energy data must be converted using the actual interval duration if the model requires average power. For a cumulative register, calculate valid differences only after resolving resets and gaps. A correct annual total does not prove that each interval was interpreted correctly.
Reconcile energy and peaks separately
Compare the prepared file’s monthly energy with relevant meter or billing totals, allowing for documented boundaries and billing-period differences. Then check peak timing and magnitude independently. Scaling the entire profile to match a monthly energy total does not establish the customer’s actual demand events.
Include a before-and-after summary that names the affected records and changes in both energy and peak values. Retain quality flags in the exported study data or a companion file. The engineer reviewing the proposal should be able to identify when a critical dispatch result depends on reconstructed information.
Make this evidence package a named deliverable in the commercial BESS procurement checklist. It belongs beside the equipment assumptions because changing the input trace can change the selected capacity without any change to the hardware.
Optional walkthrough: uploading a load profile
This official NLR Learning tutorial explains the distinction between a simulated profile and uploaded interval data in REopt Lite. It is a workflow illustration, not a substitute for the current tool’s import requirements or the quality checks above.
The official text version of the tutorial provides an accessible alternative. Neither format establishes that an individual customer’s dataset is complete or accurate.
Handover a reproducible dataset
Before requesting a quote for commercial battery storage, provide the raw file, prepared file, transformation log, quality register, meter-boundary description and model time convention. Mark the unresolved periods and explain how the sizing study treats them.
The resulting package should let another analyst reproduce the same input without guessing. That is a stronger foundation for a battery investment than a polished graph whose cleaning decisions have disappeared.