Calculating the cost of load forecast error at feeder level

EUR 50,000 to EUR 55,000 per year is the combined exposure for this example feeder. That total comes from two smaller figures: EUR 550 for each 10 MW under-forecast and EUR 350 for each 10 MW over-forecast. The calculation shows how forecast error becomes an operational cost.

Building the example

Consider a medium-voltage feeder serving a mixed residential and light-commercial load zone. Its peak demand on a typical autumn weekday is approximately 200 MW. The distribution operator buys day-ahead reserves for forecast demand and a reserve margin at a contracted standby rate. Here, that rate is approximately EUR 35/MW/day, within the range reported in REN and ENTSO-E Iberian reserve market summaries.

A 5% error on 200 MW equals 10 MW. Its procurement cost depends on both directions.

Under-forecasting: the cost of emergency procurement

If demand comes in above forecast, the operator enters the dispatch period with less generation committed than the load requires. Once the shortfall exceeds the contracted reserve margin, extra generation must be obtained at short notice, usually through the intraday balancing market or emergency reserve activation.

Prices for short-notice procurement depend heavily on market conditions. Iberian intraday balancing generation has traded at a 40 to 80% premium over day-ahead reserve contract prices during tight system margin. Using EUR 55/MW/day at the midpoint of that premium range, a 10 MW shortfall costs EUR 550 per occurrence.

The per-event amount is only part of the picture. If 20% of peak-period dispatch days require emergency procurement, and there are 250 peak-period days per year, the feeder has 50 emergency events. At EUR 550 each, that is 50 emergency events x EUR 550 = EUR 27,500 per year per feeder from under-procurement.

Over-forecasting: the cost of standby waste

Over-forecasting creates a quieter cost. The operator commits reserves for the forecast peak and its margin, then actual demand arrives lower. The standby generation is not dispatched, but the contracted capacity was still reserved and paid for.

At EUR 35/MW/day, reserving 10 MW too much costs EUR 350 per occurrence. If 30% of operating days over-procure, 250 days per year produces 75 occurrences. The result is 75 x EUR 350 = EUR 26,250 per year per feeder in standby waste.

Adding both directions of the 5% forecast error gives roughly EUR 50,000 to EUR 55,000 per year per feeder under these assumptions. Contract design, market prices, and the distribution of errors across demand levels will change the result. We are not presenting this as a universal figure; we are presenting the calculation methodology.

Why error cost does not scale linearly

The 5% case costs roughly EUR 50,000 per year per feeder. Larger errors do not scale directly. Two effects complicate the relationship.

First, emergency procurement becomes more expensive as the gap widens. A 5 MW shortfall may be coverable from spinning reserve. A 20 MW shortfall may push into balancing market at much higher prices, especially when system margin is tight. Each additional MW of under-procurement can therefore carry a higher marginal cost.

Second, the shape of the error distribution matters more than its average alone. A forecast with a heavy tail and occasional large misses is more expensive operationally than one with the same average error but a narrow distribution. Rare tail events create the most costly emergency procurement. P90 coverage rates track this, not just average error, because tail behaviour determines procurement risk rather than the median error.

How the economics change across a feeder portfolio

For an operator managing 15 to 20 feeders, forecast errors do not simply behave as independent values that can be added together. Geographically proximate feeders have correlated errors. A weather event that causes underestimation on one residential feeder can produce underestimation on nearby residential feeders at the same time, making portfolio reserve sizing more involved.

Errors can also partly cancel across a different load mix, such as residential / industrial / agricultural feeders. An agricultural feeder may underestimate during irrigation activation, while an industrial feeder may overestimate because a plant is in maintenance. The portfolio position can therefore be better than the result for its most exposed individual feeder.

This is why reserve management at portfolio level can be more useful than setting reserves feeder by feeder. Looking at the joint P90 of a 15-feeder portfolio can support a lower total reserve position than adding each feeder's P90 independently, because individual extremes do not all happen simultaneously. Feeder-level probabilistic forecasting makes this portfolio calculation possible.