accuracy of normalized power algorithm

Almost dead-on...and boy, does it hurt.

Jim

Funny you should bring this one up, Andy. I was just thinking about the question because in yesterday's fast club ride I was riding in the middle to the back of the pack and getting whipsawed pretty badly. Of course, my power was all over the place during that time. Eventually I got dropped, I eased up (at least by PE, I wasnt watching the power meter) my power smoothed right out, but my NP was virtually identical for the period before and after I "let up". It makes me suspicious that NP is underestimating when power is widely variable.

I voted for "almost always" but I was tempted to vote "occasionally overestimates by more than 5%" because I get a lot of rides with np nearly 105% of ftp and, much less frequently, just more than 105%. If I used the power at which I normally do 2X20s as ftp instead of actual 1 hr best time trial power, I'd come down more solidly in the "almost always" camp so maybe its a training specificity issue.

I voted for "almost always w/in 5%" but I was tempted to vote "occasionally overestimates by more than 5%" because I get a lot of rides on the cusp of 105% and, much less frequently, just over 105%. If I use the power that I typically generate for 2X20s as ftp instead of actual best 1 hr time trial power, I'd come down more solidly in the "almost always" camp so maybe it's a training specificity issue instead of an issue with the NP algorithm.

Andy

In the 5 hard races I have data for, they are all crits < 1hr (30-50min), so I haven't put a poll result in since they don't meet the criteria.

Having said that, the PNorms I get for these crits are 7%-9% higher than my estimated FTP at the time. Indeed, these results keep telling me to keep my FTP up, but then I find I can't sustain it when it comes time to do steady state FT effort and then I wonder if I've over estimated FTP. I am willing to consider a number of reasons for it (e.g. lower arousal levels when training vs racing; simply not fresh enough when doing training efforts vs being fresh for a race; time of day - training at 6am vs racing later in day and so on). Since the training levels are a continuum, I don't fret too much about it but it would be nice to actually be able to crank it out at the upper end of the range every once in a while in training....

Everything else is track racing, so much too short but I generally get a lower PNorm on the track than PE would suggest. In a recent hard 30min points race PNorm = FTP estimate, whereas in crits I score much higher PNorms. For me, I think that's a fixed gear efficiency/cadence thing.

Alex

Where's the choice for "it depends?" 😄

flapsupcleanup said:

It makes me suspicious that NP is underestimating when power is widely variable.

I would generally agree with that, if I'm reading the question properly. A very wildly varying race with lots of short bursts or accelerations out of corners (and subsequent periods of coasting) feels *harder* then a fast, steady ride, yet produces an NP that is slightly-to-somewhat lower.

For a steadier (but still hard) race, I feel NP is pretty darn close to what I can produce during typical training-as-testing, even in rolling terrain.

frenchyge said:

I would generally agree with that, if I'm reading the question properly. A very wildly varying race with lots of short bursts or accelerations out of corners (and subsequent periods of coasting) feels *harder* then a fast, steady ride, yet produces an NP that is slightly-to-somewhat lower.

For a steadier (but still hard) race, I feel NP is pretty darn close to what I can produce during typical training-as-testing, even in rolling terrain.

Yep, I'm with you there, but the wildly varying race not only "feels" harder, but can actually blow me up, yet right after that when the pack is disappearing my NP for the next 20 mins is the same as the 20 before the drop.

My understanding is that that's the intent of NP, to account for the wild swings.

So there are two possibilities in my mind here. One is that the NP algorithm is only accurate within a certain variability band. The other (more likely, I'm afraid) is that I am just poor in L6+ and those bursts take more out of me than the guy next to me even though our FT/wt is the same.

frenchyge said:

I would generally agree with that, if I'm reading the question properly. A very wildly varying race with lots of short bursts or accelerations out of corners (and subsequent periods of coasting) feels *harder* then a fast, steady ride, yet produces an NP that is slightly-to-somewhat lower.

For a steadier (but still hard) race, I feel NP is pretty darn close to what I can produce during typical training-as-testing, even in rolling terrain.

It's also possible that one measure doesn't entirely measure both the aerobic and anaerobic capacities. I have always viewed NP as primarily for measuring my aerobic stress versus my aerobic capacity and that my anaerobic capacity and use of that resource requires further analysis. I'm working on that problem now.

flapsupcleanup said:

Yep, I'm with you there, but the wildly varying race not only "feels" harder, but can actually blow me up, yet right after that when the pack is disappearing my NP for the next 20 mins is the same as the 20 before the drop.

My understanding is that that's the intent of NP, to account for the wild swings.

So there are two possibilities in my mind here. One is that the NP algorithm is only accurate within a certain variability band. The other (more likely, I'm afraid) is that I am just poor in L6+ and those bursts take more out of me than the guy next to me even though our FT/wt is the same.

Think AWC...

RapDaddyo said:

It's also possible that one measure doesn't entirely measure both the aerobic and anaerobic capacities. I have always viewed NP as primarily for measuring my aerobic stress versus my aerobic capacity and that my anaerobic capacity and use of that resource requires further analysis. I'm working on that problem now.

I normally don't worry about *analyzing* that portion of my fitness, though I do track it via the power profile.

I would love to see how you solve this problem, as it could have a very desirable outcome; measuring matches...

If you need anything from me, ping me and I will do what I can...

Jim

frenchyge said:

Where's the choice for "it depends?" 😄

Andy, it depends on what convention we're using to set FT (see other thread). If you settle this, I can re-run my Cyclingpeaks and answer your question very well.

FWIW, I set FT in a 20 minute test. No corrections. It helps athletes show big numbers, and then they're convinced their coach knows what he's doing 😉

Spunout said:

Andy, it depends on what convention we're using to set FT (see other thread). If you settle this, I can re-run my Cyclingpeaks and answer your question very well.

Well in that case, I say that functional threshold power should be defined as normalized power during a hard ~1 h race or workout. 🙄 😉

Seriously, in the context of this poll I'm not sure it really matters. With the exception of using 5 min power divided by 1.2, any of the various methods is going to yield an estimate for functional threshold power that is w/in ~5% of the "true" value the vast majority of the time. Thus, if people are regularly observing normalized power to be off by >5%, it suggests that the limitation lies in the algorithm, not how you define functional threshold power.

Spunout said:

FWIW, I set FT in a 20 minute test. No corrections. It helps athletes show big numbers, and then they're convinced their coach knows what he's doing 😉

Do you mean 20 min power, or 20 min power * 0.95?

EDIT: Oops, you do say "no corrections" - my bad.

flapsupcleanup said:

Yep, I'm with you there, but the wildly varying race not only "feels" harder, but can actually blow me up, yet right after that when the pack is disappearing my NP for the next 20 mins is the same as the 20 before the drop.

My understanding is that that's the intent of NP, to account for the wild swings.

So there are two possibilities in my mind here. One is that the NP algorithm is only accurate within a certain variability band. The other (more likely, I'm afraid) is that I am just poor in L6+ and those bursts take more out of me than the guy next to me even though our FT/wt is the same.

On occasion I insert a power file into a spreadsheet. One column shows each 5 second period (you could use one second or other, too) where power was higher than threshold. Also shown is how many watts each period was above threshold. Total up the watts (above threshold) and divide by some number that provides decent scaling.

For criteriums with some hard out of corner sprints and similar efforts this provided a decent picture (graph) of one aspect of the race's difficulty.

I made a graph with a bar for average power pedaling only, a bar for average total, a bar for this variability total, and average HR.

I'll vote after my first race of the season- tomorrow. 40k rolling course should give me something to use.

PhilH said:

I'll vote after my first race of the season- tomorrow. 40k rolling course should give me something to use.

Me too...I've got my first race of the season Saturday. But I'll pick the hardest hour (the last hour in a Pro/1/2 event) and post it.

I voted for occasionally overestimates but it is possible that I need to retest my FT. My 2 hr race today had an NP = FT(1hr). The first hour of it being at 1.06. So in reality I think I probably should have voted for the middle one.

Phil

Hard training crit, one hour.

With my FT logged as 280, my NP was 284; AP 250 and IF 1.025. I'd say its all good.

I won't bump it up my FT yet, but after this block I will re-test and it will be higher (I'm TTing at 296W over 15km).