Studies

Dominion Resources

Goal: 10% reduction in the load forecasting error rate.

• From May to December 2005, the load forecasting errors were costing them between $3.5 million to $4 million each month.
• Their year-to-date September 2006 mean load forecast error was ~ 2.95%, compared to their 2005 baseline of 3.43%.
  • That translated into a 14% improvement in a process once thought unimprovable.
  • The financial savings to Dominion exceeded more than $20 million in avoided purchased electric power.

In the regulated ISO Zones, even electricity Generation utilities are required to purchase electricity on the “Open Market” (at Spot Prices) when they under-forecast and thus under-commit MW’s in their day-ahead markets. Dominion Generation paid $800 Million in 2011 for these spot buys at an average cost of $50/MWHr, when they produce it for avg of only $20/MWHr. These Trading Zones cover approximately 75% of the USA, which is why Electricity Trading has become such a major industry, starting when markets were deregulated as monopolies in the mid-1990’s. With the current shift into 1,000’s of distinct price locations (Nodes) in the organized North American wholesale electric markets, load forecasts also grow in importance.

Source: http://store.isixsigma.com/compare/Project-Examples/dmaic/electric-load-forecasting-project-example#compare_tab

Maximize Energy Trading Profits w/ Accurate Load & Price Forecasts

Accurate load and price forecasting for electricity in the regional (ISO) markets can produce large profits for energy traders. This white paper describes different load and price forecasting techniques utilized today, and the use of a neural net prediction algorithm for predicting price spikes in the New England ISO region.

Source http://www.energycentral.com/download/products/Gaushell.pdf

SAIC study: Economic Significance of Improved Environmental Forecasting

A review of forecast errors and associated market prices shows that significant forecast errors occurred during periods of high market prices. This relationship suggests that marketers may be able to pinpoint key load and temperature events, and that there is a need to reduce load forecast errors.

Further investigation indicated that the weather forecast component in the model was responsible for only ~ 40% of the overall load error. A subsequent test of the 30 days with the highest load error revealed that when the load error was high, forecast error was responsible for a substantially larger percentage of the load model error. The forecast improvement was only 6.41% for the entire model run. This improvement more than doubled to 15.76% when the analysis was conducted on the 30 days with the highest error.

Source: http://www.imagingnotes.com/go/article_free.php?mp_id=93

The Economic Benefit of Incorporating Weather and Climate Forecasts into Western Energy Production Management

California ISO: Ensemble Forecast Modeling.  The cost of a demand forecast based on a temperature over-forecast can be significantly less costly than a temperature under-forecast of the same amount; specifically the cost of a +4000 MW error is about 1/4th of a –4000 MW error. Given this asymmetry in the cost curve, there is significant utility in moving away from the mean (or median).

Source http://cirrus.ucsd.edu/~pierce/calenergy/deliv4.pdf

SAIC: Deficiencies

A series of benchmarking energy sector studies showed the following deficiencies:

• Error in weather forecast data
• Inadequate incorporation of environmental data into load forecast models
• Load forecast model error
• Misapplication of load forecast output in business transactions

Source: http://www.imagingnotes.com/go/article_free.php?mp_id=93

Improving Weather and Load Forecasting For The California Independent System Operator (CAL ISO)

Case Study 1: Improving the Forecast of Delta Breeze and Determining The Economic Value.

Herein, the average cost of over forecast was approximately $150/MW. The total cost of Delta Breeze induced load forecast error, due to the .45 weather factor was associated with a seasonal cost of about $9.9 Million. Improving the temperature forecast mean error by reducing it to about .4 deg-F vs. the .45 deg-F and using a standard deviation of about 2.7 degree-F would save the Cal ISO about $2 million/season.

Source: http://cirrus.ucsd.edu/~pierce/calenergy/deliv3.pdf

Renewable Loads

A 10% reduction in MAPE forecast error can reduce ‘load following capability’ by 100’s of MW’s each hour:

A CAISO study revealed aggregate “all hours” results that compare the load following up and down MW’s calculated in each hour, with and without errors, for all hours in a season. In each case, variability contributes 19% of the total requirement, with forecast errors providing the remaining 81%. The hourly results show in which hours improvements in forecasting are likely to provide the highest benefit.

The sensitivity analysis of forecast error provides a quantitative measure of how improvements in the hour-ahead forecast (and in periods further ahead) can reduce the ramp range that an ISO will need to deploy within the hour. A 10% reduction in forecast error could result in a reduction in several hundred MW of load following capability in an upward or downward direction.

Results point to the particular hours (morning and evening ramps), where such forecast improvements have the most value.

Historically, given its variable nature, wind generation has been taken on an as-available (or “must take”) basis, and grid operators compensate by incrementing or decrementing the output of other committed generation. At low wind penetrations, such actions do not significantly affect system operations. At higher levels of wind penetration, however, forecast uncertainty becomes more challenging. So actual wind generation and the forecasted wind generation, become ever more valuable, in the hour-ahead time frame. Improvements in forecasts will facilitate renewable integration by allowing operators to ensure that the right resources are committed and on dispatch to address actual variability.

Source: https://www.caiso.com/2804/2804d036401f0.pdf