Senior decision-makers come together to connect around strategies and business trends affecting utilities.

Post

Crystal Ball Lessons in Predictive Analytics

Tao Hong's picture
  • Member since 2014
  • 8 items added with 5,459 views
  • Apr 7, 2015
  • 1994 views
ANALYTICS IS AN INTEGRAL part of this business, and whether it is because of a lack of data or expertise, the business case to invest in analytics has been difficult to nail down. But what if your company had a crystal ball for analytics? How easy would it be to make the case then? The traditional folk tale's crystal ball foretold the future: what will happen, what won't, love coming in, love going out. A business-specific foretelling, though not as lovey-dovey, could give you 100 percent accurate answers to questions about the revenue you could expect for the next five years and from what lines of business it will come. It could tell you when the next price spike in the electricity market will come and how much it will be. It could foretell tomorrow's peak demand and how many transformers will be taken down by next week's storm. It could tell you how many customers will buy electric vehicles next year.
So, forget the traditional crystal ball. Instead, how much would you pay for one that's accurate for business predictions?
The valuation for the perfect crystal ball would depend on how large the business is. The bigger the company is, the higher the potential value the crystal ball has. The valuation also depends on how good or bad are the present results of business forecasting. If you already know how to predict the future with 100 percent accuracy, then the crystal ball is worth nothing. If your forecasting ability otherwise is poor, then the crystal ball may be quite valuable. And the valuation further depends on how the crystal ball would be used. Would you use it for your entire enterprise or only in one department, such as power supply, customer services, or transmission planning? As the use of the crystal ball moves up the organization chart, its apparent value grows significantly. The effort needed to estimate its value also grows accordingly.
This list can go on and on. After you evaluate all possible factors influencing the valuation, you will end up with a problem too complex to answer effectively. Instead of pursuing the perfect valuation with endless analysis and debates, we can obtain a ballpark figure with a reasonable effort that considers only one area of business forecasting in the form of predictive analytics.
Long-term Load Forecasting
Consider a utility with a 1-gigawatt annual peak load. Assuming the utility uses long-term load forecasting for generation planning only, how much should we value the crystal ball if it could improve the company's forecast accuracy by 1 percent?
The risk of oversizing or undersizing a power plant is: 1,000 MW x 1% = 10 MW
Assuming the capital cost is $10,000/kW, the overnight capital cost is: $1,000/kW x 10 MW = $10 million
The savings of deferring $10 million in spending for one year (5 percent interest rate): $10 million - $10 million / (1 + 0.05) = $476,000 (~$500,000)
Therefore, the crystal ball can be valued at $500,000 per year for improving the forecasting accuracy of a 1-gigawatt peak utility by 1 percent.
Short-term Load and Price Forecasting
Assuming the utility uses short-term load forecast for securing energy in the day-ahead market only, how much should we value the crystal ball if it could improve the company's forecast accuracy by 1 percent? What if the crystal ball could also predict the prices in day-ahead and real-time markets?
Electricity demand is heavily driven by weather, and most market participants use similar sources of weather forecasts. When you, as a typical market participant, are under- forecasting (or over-forecasting) the day-ahead demand, most other participants are doing the same. As a result, if you buy more energy than actually needed in the day-ahead market, you then have to sell the energy in the real-time market at a lower price. If you didn't buy enough in the day-ahead market, then you have to buy energy in the real-time market at a higher price. In other words, you almost always lose money in the real-time market because of the day-ahead forecast errors.
For example, take ISO New England's demand and price data for the past decade (2005-2014). We first standardize the hourly demand so that the standardized peak load of each year is 1 gigawatt. Now ISO New England is in the same size as the 1-gigawatt peak utility previously mentioned as an example. We can calculate the cost of 1 percent demand being placed in the high-price market.
Without a price forecast, we would try to place as much demand in the day-ahead market as possible because the day-ahead market is less volatile than the real-time market. We can calculate the cost of placing 1 percent of the demand in the day-ahead market. With a price forecast, we know which market has a lower price, thus, we can place the 1 percent demand in the low-price market. We also can calculate the associated cost. Comparing this cost with the cost of placing the 1 percent demand in the high-price market, we can then obtain the savings for both cases as listed in the table.
On average, the predictive analytics crystal ball saves about $300,000 per year for improving the company's short-term load forecast accuracy by 1 percent. With the additional price forecasts, the savings can rise to $600,000 per year.
 Numbers for Back-of-the-envelope Calculations
Based on the aforementioned calculations, here are three magic numbers for your back-of-the-envelope calculations, all based on 1 percent improvement on a 1-gigawatt peak utility:
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
For a typical medium-size utility with a 5-gigawatt peak load, taking an integrated load forecasting approach to improve long-term load forecasts by 3 percent and short-term load forecasts by 1 percent, the total annual savings would be around: ($500,000 x 3 + $300,000 x 1) x 5 = $9 million.
As a ballpark estimate, this means that the actual annual savings would be most likely to fall between $4.5 million and $18 million, but not outside the range between $900,000 and $90 million.
Finding your Crystal Ball
These so-called crystal ball numbers do not reflect reality, of course. In reality, because of the stochastic nature of the world, such a crystal ball does not exist. Nevertheless, you always have access to predictive analytics, which helps forecast the future. Even if your analysis returns half of the value estimated from the crystal ball we talked about, it is a strong return on investment, particularly when multiplied across other areas of the business.
The usefulness of your forecast to prompt improved organizational efficiency and improved decision-making is mostly under your control. Forecasts can always be improved, and they are as close to a crystal ball that you can own right now. The question is: How much are you willing to invest in forecasts to make your vision of the future the most accurate you can get?
Dr. Tao Hong is Energy Production and Infrastructure Center assistant professor, graduate program director of engineering management and North Carolina Electric Membership Corporation faculty fellow of energy analytics at the University of North Carolina at Charlotte.

ANALYTICS IS AN INTEGRAL part of this business, and whether it is because of a lack of data or expertise, the business case to invest in analytics has been difficult to nail down. But what if your company had a crystal ball for analytics? How easy would it be to make the case then?

The traditional folk tale's crystal ball foretold the future: what will happen, what won't, love coming in, love going out. A business-specific foretelling, though not as lovey-dovey, could give you 100 percent accurate answers to questions about the revenue you could expect for the next five years and from what lines of business it will come. It could tell you when the next price spike in the electricity market will come and how much it will be. It could foretell tomorrow's peak demand and how many transformers will be taken down by next week's storm. It could tell you how many customers will buy electric vehicles next year.

So, forget the traditional crystal ball. Instead, how much would you pay for one that's accurate for business predictions?

The valuation for the perfect crystal ball would depend on how large the business is. The bigger the company is, the higher the potential value the crystal ball has. The valuation also depends on how good or bad are the present results of business forecasting. If you already know how to predict the future with 100 percent accuracy, then the crystal ball is worth nothing. If your forecasting ability otherwise is poor, then the crystal ball may be quite valuable. And the valuation further depends on how the crystal ball would be used. Would you use it for your entire enterprise or only in one department, such as power supply, customer services, or transmission planning? As the use of the crystal ball moves up the organization chart, its apparent value grows significantly. The effort needed to estimate its value also grows accordingly.

This list can go on and on. After you evaluate all possible factors influencing the valuation, you will end up with a problem too complex to answer effectively. Instead of pursuing the perfect valuation with endless analysis and debates, we can obtain a ballpark figure with a reasonable effort that considers only one area of business forecasting in the form of predictive analytics.

Long-term Load Forecasting

Consider a utility with a 1-gigawatt annual peak load. Assuming the utility uses long-term load forecasting for generation planning only, how much should we value the crystal ball if it could improve the company's forecast accuracy by 1 percent?

 

  • The risk of oversizing or undersizing a power plant is:
    1,000 MW x 1% = 10 MW
  • Assuming the capital cost is $10,000/kW, the overnight capital cost is:
    $1,000/kW x 10 MW = $10 million
  • The savings of deferring $10 million in spending for one year (5 percent interest rate): $10 million - $10 million / (1 + 0.05) = $476,000 (~$500,000)


Therefore, the crystal ball can be valued at $500,000 per year for improving the forecasting accuracy of a 1-gigawatt peak utility by 1 percent.

 

Short-term Load and Price Forecasting

Assuming the utility uses short-term load forecast for securing energy in the day-ahead market only, how much should we value the crystal ball if it could improve the company's forecast accuracy by 1 percent? What if the crystal ball could also predict the prices in day-ahead and real-time markets?

Electricity demand is heavily driven by weather, and most market participants use similar sources of weather forecasts. When you, as a typical market participant, are under- forecasting (or over-forecasting) the day-ahead demand, most other participants are doing the same. As a result, if you buy more energy than actually needed in the day-ahead market, you then have to sell the energy in the real-time market at a lower price. If you didn't buy enough in the day-ahead market, then you have to buy energy in the real-time market at a higher price. In other words, you almost always lose money in the real-time market because of the day-ahead forecast errors.

For example, take ISO New England's demand and price data for the past decade (2005-2014). We first standardize the hourly demand so that the standardized peak load of each year is 1 gigawatt. Now ISO New England is in the same size as the 1-gigawatt peak utility previously mentioned as an example. We can calculate the cost of 1 percent demand being placed in the high-price market.

Without a price forecast, we would try to place as much demand in the day-ahead market as possible because the day-ahead market is less volatile than the real-time market. We can calculate the cost of placing 1 percent of the demand in the day-ahead market. With a price forecast, we know which market has a lower price, thus, we can place the 1 percent demand in the low-price market. We also can calculate the associated cost. Comparing this cost with the cost of placing the 1 percent demand in the high-price market, we can then obtain the savings for both cases as listed in the table.

On average, the predictive analytics crystal ball saves about $300,000 per year for improving the company's short-term load forecast accuracy by 1 percent. With the additional price forecasts, the savings can rise to $600,000 per year.

Numbers for Back-of-the-envelope Calculations

Based on the aforementioned calculations, here are three magic numbers for your back-of-the-envelope calculations, all based on 1 percent improvement on a 1-gigawatt peak utility:

 

  • 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

 

For a typical medium-size utility with a 5-gigawatt peak load, taking an integrated load forecasting approach to improve long-term load forecasts by 3 percent and short-term load forecasts by 1 percent, the total annual savings would be around: ($500,000 x 3 + $300,000 x 1) x 5 = $9 million.

As a ballpark estimate, this means that the actual annual savings would be most likely to fall between $4.5 million and $18 million, but not outside the range between $900,000 and $90 million.

Finding your Crystal Ball

These so-called crystal ball numbers do not reflect reality, of course. In reality, because of the stochastic nature of the world, such a crystal ball does not exist. Nevertheless, you always have access to predictive analytics, which helps forecast the future. Even if your analysis returns half of the value estimated from the crystal ball we talked about, it is a strong return on investment, particularly when multiplied across other areas of the business.

The usefulness of your forecast to prompt improved organizational efficiency and improved decision-making is mostly under your control. Forecasts can always be improved, and they are as close to a crystal ball that you can own right now. The question is: How much are you willing to invest in forecasts to make your vision of the future the most accurate you can get?

Dr. Tao Hong is Energy Production and Infrastructure Center assistant professor, graduate program director of engineering management and North Carolina Electric Membership Corporation faculty fellow of energy analytics at the University of North Carolina at Charlotte.

Discussions

No discussions yet. Start a discussion below.

Tao Hong's picture
Thank Tao for the Post!
Energy Central contributors share their experience and insights for the benefit of other Members (like you). Please show them your appreciation by leaving a comment, 'liking' this post, or following this Member.
More posts from this member

Get Published - Build a Following

The Energy Central Power Industry Network is based on one core idea - power industry professionals helping each other and advancing the industry by sharing and learning from each other.

If you have an experience or insight to share or have learned something from a conference or seminar, your peers and colleagues on Energy Central want to hear about it. It's also easy to share a link to an article you've liked or an industry resource that you think would be helpful.

                 Learn more about posting on Energy Central »