Thursday, April 9, 2015

Crystal Ball Lessons in Predictive Analytics

For a long time, I have had the idea of writing an article about "how much benefit are we getting from reducing load forecast errors". A few months ago I got a request from EnergyBiz to contribute an article. I thought this "valuation" topic would be a good fit. So here is Crystal Ball Lessons in Predictive Analytics.

Since load forecasts are being used by many departments of a utility, it is very hard to quantify the benefits of improving it. Most people use some rough numbers such as "millions of dollars of savings". A few academic papers studied the savings on some specific utility applications through computer simulations. I would like to find a balance between these two extremes for the primary audience of EnergyBiz, the executives in the utility industry. I would like to derive the savings through simple examples and finally offer some rule-of-thumb numbers for back-of-the-envelope calculations.

In summary, here are the ballpark savings from 1% reduction (i.e., reducing the mean absolute percentage error from 4% to 3%) in forecast error for a utility with 1GW peak:
  • Long-term load forecasting: $500,000 per year
  • Short-term load forecasting: $300,000 per year
  • Short-term load and price forecasting: $600,000 per year
The web version of the paper is available HERE. You can also download the full PDF of the issue HERE.

Tao Hong, "Crystal Ball Lessons in Predictive Analytics", EnergyBiz, pp. 35-37, Spring, 2015

Back to Error Analysis in Load Forecasting

1 comment:

  1. Your approach is interesting and has merit, but not my preference. Embedded in your analysis is the notion that load is served from generation. Instead, I assume that load is served from LMP. I formulate the load server problem as a trade, akin to a short option straddle. I have called it the "Load Server Strangle" over the years. Straddles and strangles are similar trade positions.

    How does the LSE Strangle work? Assuming no special skill in weather forecasting, cost is minimized when the LSE buys exactly correctly, which means that their weather forecast was dead center. To the extent that the weather forecast missed, purchases missed the targeted load, either too much or too little. When overbought, the LSE sells the excess into a declining market. DA>RT. When underbought, the LSE buys in a rising market DA<RT. The LSE is disproportionately on the wrong side of those trades.

    The LSE task, then, is to protect the option premium, defined in retail power as "headroom", through best practices trading. The LSE benefits when their weather forecast improves. How much? When formulated as a series of trades rather than as a capital investment, the value-added of an improved weather forecast becomes: what is the improvement in financial performance given a specified reduction in the weather forecast error?


Note that you may link to your LinkedIn profile if you choose Name/URL option.