How is load forecasting profitable to power companies

Question

Let's suppose that a power company is using a software model 'X' to do load forecasting which can predict 24 hour ahead load with a MAPE of 5%. Now suppose that I've created a model 'Y' that reduces the MAPE to 1%. So, my question is how much profit will the company make with reduction of MAPE by 4%. How to do the quantitative analysis ? I want to do it because the pricing of the model 'Y' would be based on the profit made by the company

Answer 1

Unfortunately the profit that the utility makes is dependent on many different factors, most of them unquantifiable by you, some of them unquantifiable by the utility company.

They will have a contract to supply power 24/7, with big penalties if they fail (these might be financial, written into the supply contract, or loss of reputation and inability to make future contracts).

In order to feel comfortable themselves that they are going to be able to supply all the needed power in the face of unknowable variations in demand, they run extra generating capacity at less than full power, or at zero power. Even at zero power, it still costs a considerable amount. They less extra capacity they are able to run, the better the profit. If a thermal plant takes 2 hours to get to full capacity, then this defines a critical time horizon for them. One of the benefits of the pumped storage station at Dinorwig is that it could provide full power from stationary turbines in 75 seconds, and from ready in 16 sesconds.

They will tend to be conservative. Running a little more spare capacity only costs a bit of money now. Running too little risks trashing their business.

You would need to know the appetite for risk of the people making the scheduling decisions, the penalties in their contract, the other possibilities they have of covering outages like importing power from a rival, to understand the costs.

Part of their thinking will be the proven reliability of the model. You might be doing well if they'd agree to try your model 'Y' for nothing for a year to see how well it performed under their conditions, rather than taking money off them for an unproven model.

Answer 2

I know nothing about that part of the business, but just thinking about it for a few seconds, it seems to me that the primary use of short-term predictions is to inform the scheduling of fuel deliveries and equipment maintenance. I can't imagine that a 4% reduction in prediction error would have a huge impact on the overall profit margin of a producer.

Longer-term predictions would inform the decisions around investing in new equipment and/or retiring old equipment. There would be a more direct effect on operating efficiency and profit margins. But of course, making long-term predictions in a chaotic process (just like the weather) is pretty much impossible.