xx October 20XX

Method Report: Typical Load Curves

Distribution Company RTP 20XX

Summary

The pipeline transforms the measurement campaign into 24-hour typical load curves, by class and consumption range, for weekday, Saturday and Sunday or holiday. The curves are output in the aggregation model layout, with the PS factor (Saturday relative to weekday) and the PD factor (Sunday relative to weekday), and adjusted to the annual energy of the population in BDGD.

The method has six stages: data reading and cleaning, remaining curves, representative curve for each point, grouping by shape, organization in the model layout and energy adjustment.

This document describes the method as applied to the DISTRIBUTION COMPANY RTP 20XX study, with the parameters actually used. The study results are in the results report.

1. Starting Data

• Measurements: 31 measurement files in the Measurements folder, one per class and range (or per circuit), with one tab per measurement point.
• Layout: each 15-minute value appears in three 5-minute rows; the program takes the average of the trio and converts to kW. Each tab brings the DU, DS and DD marks of the campaign in row 293.
• Holidays: 7 dates in holidays.csv (from 07/09/2025 to 01/01/20XX), with national, state and municipal coverage. Saturday and holiday have their own day types: Sunday and holiday count as DD.
• Days of the year: the year of the cycle (20XX) has 252 weekdays, 52 Saturdays and 61 Sundays and holidays. These numbers weight the weekday, Saturday and Sunday in the annual energy (Stage 6).
• BDGD: population and energy for each range come from ANEEL's stratification panel in the BDGD folder (Module 2, preferred; the old panel is used when Module 2 does not exist). Population can be entered manually in parametros_lote.csv.
• Loads and circuits: 26 files treated as load and 5 as circuit (type column in parametros_lote.csv). The circuit receives the same treatment as stages 1 to 5; the energy adjustment of Stage 6 applies only to load.

2. Stage 1: Reading and Elimination Rules

Each day of each point is tested by the rules below. The day is eliminated by the first rule that applies to it, and each rule has its count recorded in the batch summary.

R6 runs before processing, reads all files in the study and compares the tabs. It recognizes five situations: identical curve in another file (1 day is enough); same shape at another scale (3 or more days); nearly identical curve, with correlation of 0.9999 or higher (5 or more days); day repeated in the same tab (2 or more); and day identical to a day from another tab that had already fallen in R3 or R4, in which case R5 and the first comparison no longer saw the pair. A sixth verification, of suspicious header dates, only issues an alert. The marked pairs and days are in Results\R6_detection.

When the meter records energy per interval instead of power, the value comes out four times smaller. With the consumption range provided, the program decides, point by point, which hypothesis (no correction or multiplication by 4) leaves the average consumption closer to the range. The decisions are in scale_correction.csv.

3. Stages 2 and 3: Remaining Curves and Representative Curve

Stage 2 saves the 15-minute curves that passed the rules, with day type and original mark. In Stage 3, the representative curve for each point and day type is the average of all remaining days of that type, with DU, DS and DD separate. The average is used because the circuit curve is the sum of the loads, and only the average is additive.

4. Stage 4: Grouping

Only weekday curves enter, one per point, by shape: the 24 hourly averages divided by the peak. The algorithm is fuzzy c-means (fuzzification exponent m = 1.25, 20 initializations). For each number of groups k, from 2 to 8, the prediction strength of the groups is compared with that of data without groups, of the same mean and covariance (25 draws, 95th percentile).

• k is accepted when the prediction strength exceeds the limit of data without groups, the smallest group has at least 5% of the points and at least 5 points;
• evidence is strong when the margin over the limit is 0.10 or more, moderate from 0.03 onwards and weak below that;
• with fewer than 30 weekday curves, grouping is not tested and the range gets a single curve, because the sample does not allow knowing whether groups exist;
• in the census (the entire measured universe), each point becomes a Typical Consumer.

The Stage 4 spreadsheet also brings PRODIST sampling (Module 2, items 46 and 47): average CV of the 96 points and sampling error with 95% confidence (z = 1.96), against the 20% limit per stratum, with finite population correction when population is known. The result is informative and applies to the weekday; PRODIST sets the error per stratum, not per day type.

5. Stage 5: Organization in the Aggregation Model Layout

Stage 4 is reorganized in the Aggregated Curves tab: Hour column with 24 hours and one Typical Consumer per column (one per group; one per point, in the census; the entire range, in the single curve). The hourly curve for a weekday, in MW, is the average of the four 15-minute intervals. Two final rows bring the factors:

• PS: Saturday energy divided by weekday energy;
• PD: Sunday energy (and holiday) divided by weekday energy.

Both factors are calculated at the 96 15-minute points. At the end of the batch, Stage 5 of all files are copied to Typical Curves, in Loads or Circuits.

In injection series (distributed generation), the hourly curve of Stage 5 is saved with a negative sign, because it is energy entering the grid. PS and PD remain positive. The option is in config.py (INJECTION_NEGATIVE_SIGN).

6. Stage 6: Adjustment to BDGD Energy

Applies only to loads. Stage 6 adjusts the curves from Stage 5 to the annual energy of the range in BDGD, without changing the shape. The calculation has three steps:

• Measured energy: the annual energy per consumer is the sum of weekday × (DU + DS × PS + DD × PD), with DU, DS and DD the days of the year of the cycle (252 weekdays, 52 Saturdays and 61 Sundays and holidays in 20XX);
• Weight and participation: the weight of each Typical Consumer is the number of points in the group times the annual energy per consumer. The participation of the type is its weight divided by the sum of weights, and the energy of the type is the energy of the range in BDGD times that participation;
• Adjusted curve: the adjusted weekday sum is the energy of the type divided by (DU + DS × PS + DD × PD). Thus, when the aggregation model applies PS and PD to the curve, the energy of the range closes without counting PS and PD twice.

The adjustment factor (population energy divided by measured curve energy) shows how much the sample differs from the population. The program alerts when it falls outside the range of 0.5 to 2.0. The energy check compares, for each group (B1, B2, B3 and MT), the energy of the saved curves with that of BDGD, and the result is in Final_Check.txt.

In the population and energy of MT, Module 2 panel classifies the unit by demand (maximum measured). In low voltage, the consumption range in kWh per month is used.

Each group also receives a file of grouped loads (B1, B2, B3 = commercial, industrial and public power, and MT), with all columns from Stage 6 of its ranges side by side, plus the source map of each column.

7. MMGD Measurements

This study uses measurements of units with micro and mini distributed generation from the study ..\DISTRIBUTION_COMPANY_MMGD_RTP_20XX, which is processed first. The option, activated in config.py (ADJUST_MMGD_STAGE6), alters Stage 6:

• units with distributed generation ("With DG" in Module 2 panel) are removed from each range of B1, B2 and commercial, in number of units and in energy;
• each segment gains an MMGD column, with the curve measured in those units (residential for B1, rural for B2, commercial for B3), adjusted to the total energy of the units with DG in the segment;
• the sum of the ranges and the MMGD column closes with the total energy of the segment in BDGD. Industrial and public service distributed generation is not deducted;
• the BDGD energy for units with DG is taken as the energy consumed from the grid. This point is pending confirmation with the full BDGD dictionary.

When the distribution company has MMGD measurements, the work has two commands, in this order: the MMGD study first and the main study second. Without MMGD measurements, Stage 6 runs as before.

8. Checks and Limitations

• Checks: the energy of the curves from Stage 6 must close with that of BDGD in each group, with tolerance of 0.01 MWh (Final_Check.txt);
• the injection adjustment must close with the reported annual energy;
• automatic tests in the tests folder check energy closure and study situations with and without MMGD.

Method limitations, for those using the curves:

• the R6 thresholds were chosen with the DISTRIBUTION COMPANY campaign (RTP 20XX), which gave rise to the method, and were not calibrated with another database. In a new study, it is advisable to examine the pairs in R6_detection before accepting the result;
• the scale correction by 4 is an inference from the consumption range, not information from the meter;
• sampling error is informative and uses the average CV of the 96 points;
• where the sample is small, the single curve of the range does not show groups that may exist;
• energy adjustment preserves the measured shape, and energy comes from BDGD population.

9. Study Parameters

* Illustrative example prepared by TR Soluções. Distribution Company RTP 20XX is a fictitious company.