Losing faith in normalized power...



JohnMeyers

New Member
Aug 12, 2005
75
0
0
Is anyone starting to think that the algorithm for normalized power needs to be refined/complicated? I have recently done a couple of rides that seem to be massively overestimated in terms of normalized power, they are attached.

My primary concern is that so much is based upon the normalized power algorithm: IF, TSS, ATL, CTL, TSB... It seems to me that If you happen to like to do workouts with brief intense efforts, that any graphs involving these parameters are skewed and innacurate. For example, I did a ride that I expected to have a NP of about 225 (11.22.06). This (for me) correlates to about 93 TSS. The ride produced a NP of 283, or 147 TSS. That's a pretty big difference in TSS (>30%). If everyone was underestimating their training stress by 30%, it could lead to problems.

I think that the NP algorithm is very good in that a 1 hr. hard, variable effort produces a NP that is very rarely significantly above one's FT.

However, easy 2, 3, 4 hour long rides interspersed with bouts of hard efforts seem to trick the NP algorithm into grossly overestimating training load.

I'm not saying I know how to fix it, nor am I saying that I think that NP is "bad", I just think that it might need some work. Does anyone else notice this?

One could say to me: "John, you just suck at estimating your effort levels on rides" or, better yet, "John, you just have wicked AWC, and you don't feel these efforts like the rest of the world does". The honest truth is, I don't think either of these statements would be right.

Comments?

http://www.johnmeyers.info/files/Meyers_John_10_27_2006.wko
http://www.johnmeyers.info/files/Meyers_John_11_22_2006.wko

The first ride was on a trainer, the second, a ride with a group of buddies in which we screwed around all day by randomly attacking eachother :eek:.
 
JohnMeyers said:
However, easy 2, 3, 4 hour long rides interspersed with bouts of hard efforts seem to trick the NP algorithm into grossly overestimating training load.
Well, does it matter weather the TSS of some ride is 100 or 150? If your next day training session is e.g. 3*20 min at L4 or whatsoever and you can make your scheduled workout, you just do it?

I see TSS just a number reflecting, well, training stress score, and not a figure preventing or demanding anything specific for the next workouts.
 
JohnMeyers said:
For example, I did a ride that I expected to have a NP of about 225 (11.22.06). This (for me) correlates to about 93 TSS. The ride produced a NP of 283, or 147 TSS. That's a pretty big difference in TSS (>30%).
I'm not sure I understand what you mean here. What was your expectation of 225w based on? By my reading, all it says to me is that you simply rode harder than you had expected to. :confused:
 
JohnMeyers said:
I have recently done a couple of rides that seem to be massively overestimated in terms of normalized power, they are attached. .... However, easy 2, 3, 4 hour long rides interspersed with bouts of hard efforts seem to trick the NP algorithm into grossly overestimating training load....
(Read post 1-13)
http://www.cyclingforums.com/t-371076-15-1.html

I didn't argue with anyone in this thread about the presumed NP tendency to overestimate when there are very low level segments. But I'm still wondering about this fact.

I'm not sure it is overestimating those segment by *that much*. Bare in mind that the purpose of NP is to simulate, or at least to inspire itself from real physiological responses to what's happening during those 3hour rides.

New analytical tools are actually being developped at this moment. RapDaddyo is writting some. I'm writting some as well.

I'm sure this issue (if it's an issue) will be addressed one way or the other by one or more of these tools.
 
frenchyge said:
I'm not sure I understand what you mean here. What was your expectation of 225w based on?

That was the question that immediately came to me.
 
SolarEnergy said:
New analytical tools are actually being developped at this moment. RapDaddyo is writting some. I'm writting some as well.
Actually, I am doing nothing with the NP algorithm other than exploiting it for many purposes (e.g., TT pacing). In fact, my work is so dependent on NP that I have done a lot of work to verify its validity for me personally. I have found the best way to do this is to do two rides at the same NP, one at constant power and another at variable power. It is important that the targeted NP is close to one's maximum sustainable power for the given duration (whether it's 20mins or 60mins or whatever), because you want to feel equally "spent" at the end of each effort. Other than the fact that it took me awhile to mentally adjust to highly variable power rides and the fact that some rides bring into play other variables such as AWC recovery, my tests have increased my confidence in NP. I was a bit surprised, because I think I have an almost ideal MP/duration profile to be able to ride an "NP buster," given that I have a relatively high AWC relative to FTP.

Riding a highly variable power management strategy lies at the very heart of my TT pacing work and the computation of alternative strategies requires consistent planning parameters. An NP for the estimated duration of the ride is one of my key mathematical modeling parameters, along with the MP/duration curve.

So, if someone thinks they can come up with a better measure for the intensity of a variable power effort, knock yourself out. I have better uses for my time such as modeling matches.
 
RapDaddyo said:
I am doing nothing with the NP algorithm other than exploiting it for many purposes (e.g., TT pacing).
You have developed a way to parse a ride file, split it into segments. Next, it'll be up to the OP to use it, then substract all L1 segments (in between), and recompile a custom PMC based on that.

Or it'll be up to anyone else inspired by your works to process that automatically. That's just an example.

Other than the fact that it took me awhile to mentally adjust to highly variable power rides and the fact that some rides bring into play other variables such as AWC recovery, my tests have increased my confidence in NP.
The reason why you missed the bug, is that your TT strategies certainly don't involve riding a segment at < 100 w for 10 or 15 minutes during a 2hour effort.

But again, that doesn't change anything to the fact that some might find some more potentially interesting uses of your tools.
RapDaddyo said:
if someone thinks they can come up with a better measure for the intensity of a variable power effort, knock yourself out. I have better uses for my time such as modeling matches.
Well that won't be a difficult sale for me. I too love NP.

Moreover, I reiterate that NP's job is to implement physiological responses in the best way possible. Those physiological responses sometimes have a funny impact on heart rate, breathing rate, basic metabolical rate. Therefore I am more than willing to forgive NP's *presumed* weaknesses.
 
frenchyge said:
I'm not sure I understand what you mean here. What was your expectation of 225w based on? By my reading, all it says to me is that you simply rode harder than you had expected to. :confused:
What I am saying is this: Had I done those rides at the expected steady state power derived by the NP algorithm, I would have been exhausted by the end. Rather, I felt fresh. Thus, I don't think they represent accurate training stress scores relative to say, an hour at threshold (steady state) which would score me 100TSS, but leave me significantly more tired than I was on the 22nd (where I had a TSS of 150).

Oh, and the 225 number was based on "feel", which of course is not entirely scientific. But, I don't need science to tell me that I would have felt a lot more tired had I kept a steady state of 283.
 
JohnMeyers said:
Oh, and the 225 number was based on "feel", which of course is not entirely scientific. But, I don't need science to tell me that I would have felt a lot more tired had I kept a steady state of 283.
What would you estimate your FTP to be at the time you did those rides?

Peak 5s = 1318w
Peak 10s = 1140w
Peak 20s = 1043w
Peak 30s = 795w


I won't tell that these figures are unlikely. But they would require from you, a certain level of effort.
Have you ever experimented file download issues?
 
JohnMeyers said:
Oh, and the 225 number was based on "feel", which of course is not entirely scientific. But, I don't need science to tell me that I would have felt a lot more tired had I kept a steady state of 283.
Normalized power is not average power, and it is definitely not steady state power.
 
JohnMeyers said:
What I am saying is this: Had I done those rides at the expected steady state power derived by the NP algorithm, I would have been exhausted by the end. Rather, I felt fresh.
Ok, I think that helps. So, I'm guessing 225w was the AP for the ride and 283w the NP (sorry, I know you attached the files, but I dislike importing other people's workout files into my calendar for viewing). Would you rate the perceived exertion for that ride as equivalent to a 225w steady ride, or somewhere in between 225w & 283w?

Regarding freshness and riding by feel, I personally believe it's easier to gauge our efforts by feel when we are riding very near, or above, FTP. For me, it's easy to believe I'm fresh when riding around at low powers with an occasional burst of effort, until suddenly the legs just refuse to respond. It's quite a different sensation than grinding away at steady power, and yet the end result (eventual failure by fatigue) is the same. YMMV.
 
SolarEnergy said:
The reason why you missed the bug, is that your TT strategies certainly don't involve riding a segment at < 100 w for 10 or 15 minutes during a 2hour effort.

Point of clarification: the 'RATSS' effect to which Rick alluded concerns the calculation of TSS, not normalized power.
 
It's very possible that NP is significantly "off". That's not to say that there's any measure better than it right now. Having moved from an area with no steady ground my AP/NP spread would be around 40 watts for my longer rides. Then trying to repeat those NPs as AP leads me to believe that there not equivalent. With the right combination of efforts you can do a moderate ride that would not be impossible to replicate at steady state.

I did a L6 ride where the TSS for the intervals added up to 41- (the rest period added up to 22) but the total TSS was 111 for 71minutes. Many rides with long periods of low power will give me an extra 50 TSS per ride. This applies to L6 intervals, group rides, races, and rides with long descents. The biggest issue is when you combine these rides with rides where NP/TSS hits it pretty accurately. Ie A TSS isn't a TSS. Remember, NP is quoted as being accurate to within 5% at 1hr. Then 5% gets a lot larger when it's squared. Id think that that's probably its most accurate time point. Shorter and AWC has too much of an influence (to the point that Andy wanted CP to "hide" NP up to 20min). I already explained my thoughts on longer rides above.

Until it's either validated scientifically or something more refined comes out than its still the best we have but a lot of people believe in it a bit too literally.
 
frenchyge said:
I dislike importing other people's workout files into my calendar for viewing
You don't have to import it, you can just double-click the file and it'll open in Cyclingpeaks (assuming you've got the associations set on your computer). The nice thing there is you can see what the originator has set for FTP, weight, etc.

Alternately, do you have a "Joe Athlete" profile in your copy of CPS? You can import into there and just add notes as to who it is in the notes. I'll do this for people who I have 1-2 files for, but don't want to bother creating a new profile.
 
Well TSS is NP (actually IF which is just NP as a ratio)^2X duration . So since NP is the portion that squared than that says that that's probably the component in TSS that's off when RATSS is involved. It can only be either the IF side, the squaring factor, or the time side.
 
gvanwagner said:
Well TSS is NP (actually IF which is just NP as a ratio)^2X duration . So since NP is the portion that squared than that says that that's probably the component in TSS that's off when RATSS is involved. It can only be either the IF side, the squaring factor, or the time side.

Again, the specific effect which Rick discussed pertains only to TSS.
 
gvanwagner said:
I did a L6 ride where the TSS for the intervals added up to 41- (the rest period added up to 22) but the total TSS was 111 for 71minutes.

TSS is a global measure, and is not, and was never designed to be, additive across segments.

gvanwagner said:
Id think that that's probably its most accurate time point. Shorter and AWC has too much of an influence (to the point that Andy wanted CP to "hide" NP up to 20min). I already explained my thoughts on longer rides above.

Based on what I've seen, I would hypothesize that the "confidence limits" approach a minimum of ~5% after ~1 h, i.e., things get neither significantly better nor significantly worse after that point.

gvanwagner said:
Until it's either validated scientifically

Lindsay Edwards has actually undertaken just such a study - you can view some of his preliminary results here: http://www.endurancesportscience.com/files06/files/10-Edwards and Coggan.pdf
 
acoggan said:
Point of clarification: the 'RATSS' effect to which Rick alluded concerns the calculation of TSS, not normalized power.
Not too sure what RATSS is or I haven't followed its evolution in time. But I can confirm that based on the testing I've done myself, it does affect NP indeed (in fact, NP is taking a huge hit), then TSS.
 
It seems to me that if NP is inaccurate or misleading for a particular athlete for a particular duration, then he/she should be able to consistently do variable power efforts at that duration at a significantly different NP than he can ride constant power efforts under the identical conditions. The most common assertion is that NP overstates one's sustainable power for a given duration. But, these are almost always isolated rides, cherry picked from a bunch of workouts. As such, they are essentially worthless as evidence of anything.