Skip to main content

Best Practices for Model Calibration

Model calibration is a critical step for any simulation study. The primary objective is model validation—ensuring that the simulation accurately performs according to historical market rules for scheduling, pricing, and dispatch (SPD).

Understanding where, when, and why simulated prices differ from historical prices is essential to prevent over-calibrating your model to administrative anomalies.

1. Managing Out-of-Merit Operations & Deratings​

Out-of-merit operations—where a unit's generation does not follow its market bid—are major factors that disrupt calibration. Use the HistDBErr.csv file alongside iPoolol's built-in iPool logic to identify these events.

  • Outages Despite Low Bids: This occurs when a bid is highly competitive (e.g., negative, zero, or < 100 PHP) but the station generates no power. The iPoolol algorithm detects these anomalies (specifically targeting stations with < 100% capacity factors to exclude must-run/peaking plants) and logs them into the Ev_HST.csv file.
  • Link Deratings: Partial deratings are difficult to spot manually. iPool automatically detects these by identifying unusual flat-lining patterns in link MW flows (since normal flows fluctuate) and writes these events to Ev_HST.csv.
  • No-Bid Generation: Generating power without an active bid is non-compliant behavior, often seen with newly commissioning units. While iPool attempts to capture this by creating a flat OM_HST.bid file, this can inaccurately depress simulated market prices. Action: Manually verify these instances and consider setting the affected site to Must-Run instead.

2. Tuning Reserve Parameters​

Calibrating reserve requirements is highly dynamic across the day. You must determine the optimal reserve parameter (-pcsr x, which scales from 0.1 to 100).

  • Establish Price Bounds: Your historical target price should naturally fall within the simulated bounds of -spin 0 (minimum reserve) and -spin 1 (maximum reserve).
  • Resolution Sensitivity: Calibration logic must be adjusted depending on your dispatch resolution. A 5-minute dispatch behaves differently than an hourly dispatch due to the smoothing effects of hourly averaging.

3. Identifying Price Substitution Methodology (PSM)​

Price Substitution Methodology (PSM) is a form of administered pricing used by operators during severe intra-regional congestion. Because iPoolol strictly models standard Price Determination Methodology (PDM) rules, it cannot natively model PSM operator decisions.

  • Spot Unmodelable Prices: PSM events are indicated by historically high prices that cannot be replicated in a simulated backcast.
  • Analyze the RTD File: Look for massive discrepancies between the Locational Marginal Price (LMP) and the Loss Factor System Marginal Price (LFP-SMP). Under normal conditions, these values should be very close, differing only by the loss factor multiplier.
  • Analyze the MP File: Check the Marginal Price (MP) file to identify the exact marginal unit setting the anomalous price during that interval (e.g., a specific battery or peaking plant).
  • Avoid Over-Calibration: Recognize that iPool will naturally simulate much lower prices (e.g., 5,000 PHP) during a real-world PSM event. Do not forcefully tweak your model parameters to match these artificial administrative price spikes.