## NP v's Average Power

A light bike doesn't replace good fitness.

Moderator: Moderator Team

dynaserve
Posts: 272
Joined: Fri Jul 18, 2008 8:25 pm
Location: UK
Tinea Pedis wrote:- what power meter are you using
- what head unit
- are you taking NP off the head unit or after letting GC or TP crunch it?

I know the conversation has moved on but its Powertap + Garmin + Poweragent.

strobbekoen
Posts: 4446
Joined: Wed May 18, 2005 6:24 pm
Location: BELGIUM
Hey Guys,

Normalized power is a way of calibrating average power with intensity.

The way it is calculated is as follows
Imagine your workout as a timeline with X number of samples.
Then Imagine a 30 second time window sliding through the samples one at a time.
For simple case with a sample at each second, it will be seconds 1..30, 2..31, 3..32 etc
For each 30 seconds window, the average power is calculated. This number is then raised to the power of 4. Let's call this number Y.
As the window slides through the workout samples, each Y is added up and the number of 30 second windows is counted as N.
Normalized power is then the sum of all Y divided by the number of 30 seconds windows, raised to the reverse power of 4 (1/4) = (sum(Y) / N)^(1/4)

Your normalized power will be higher than your average power if your power profile has higher intensity / variations in power. if you ride at fairly constant power, there will be little or no difference.

Since normalized power smoothes out power, it's not useful over shorter periods. We only display it in our software if duration of the view is >= 20 minutes.

Intensity Factor = Normalized power / Power at anaerobic threshold.
Riding at your threshold yields a Intensity Factor = 1

Stress Factor = (IF)^2 * DurationInHours * 100
A Stress Factor (TSS) of 100 means one hour riding at normalized power = anaerobic power. The higher the intensity factor and duration, the higher your TSS is.

It's definitely not useless if you know how to interprete the numbers.

Rick
Posts: 2001
Joined: Sat Aug 29, 2009 4:30 pm
WMW wrote:I agree. I've tried to goad the physiology experts into coming up with a better normalizer, but there was no interest.

I am certainly no power expert, but from what I understand, NP is an estimate; so as an estimate, it needs to make some assumptions and those assumptions would be based on how it relates to an average individual. So there is not necessarily a "better" number, because different people, having inherently different critical power curves, will react differently to various levels of power fluctuation.

Or stated another way, even if they did creat a calculation that was absolutely perfect when applied to a statistically average person, it would still probably overestimate or underestimate a little when applied to each individual. So the important questions are "why do you want to know NP" and "What are you going to do with it?"

As you pointed out, there already are different estimators of "equivalent power output", so if there is one that works better for you, just freely use it.

WMW
in the industry
Posts: 855
Joined: Sun Jul 29, 2012 2:59 pm
Location: Ruidoso, NM
With a more sophisticated model, you'd be able to tweak some variables to customize it to yourself. It is going to be a moving target anyway because your CP numbers change... but NP is a lot cruder than it needs to be IMO.
formerly rruff...

DanW
Posts: 1037
Joined: Fri May 02, 2008 5:39 pm
Location: Here, there and everywhere
but NP is a lot cruder than it needs to be IMO.

I am very new to training with power but the common analysis and interpretation approaches for power waveforms appears to leave a lot to be desired. Applying various calculations, normalisations and smoothing operations without any regard for the waveform itself just goes against everything that appears logical. As I say, I am new to power on the bike but there appears to be a lot of improvements to be made.

Tinea Pedis
Posts: 8425
Joined: Sat Mar 14, 2009 6:08 am
Location: Geelong
Contact:
No system is perfect, in any walk of life.

You just need to be aware of its short falls.

DanW
Posts: 1037
Joined: Fri May 02, 2008 5:39 pm
Location: Here, there and everywhere
It is more that examining the signal and then selecting the correct signal processing based upon the signal itself isn't too difficult and a much more decent approximation using this approach could be written into a bit of software quite easily, or at least different definitions of NP could be selected for different disciplines, just as you might display a 10s or 5s moving average perhaps during a race rather than 30s.

I hadn't really appreciated that NP affects IF and TSS calculations too and you'd imagine that something like MTB in particular with lots of very short sharp bursts and its actual demands will be poorly represented by a 30s moving average. The way the TrainingPeaks article on the subject talks about it comes from the point of view that variability in the signal isn't the best representation of the actual physiological demands whereas I guess in some cases the variability might be crucial data. I guess it is just the one size fits all approach that seems odd when cycling takes many forms. I can understand how everything has originally developed but there seems to be a lot of scope for variations by looking at the actual data itself and selecting appropriate signal processing...

WMW
in the industry
Posts: 855
Joined: Sun Jul 29, 2012 2:59 pm
Location: Ruidoso, NM
I don't see it as a "one size fits all" issue. It doesn't fit anyone accurately, though you may get passably close for the efforts you care about.

I also don't see a need for different formulations for different disciplines. Different *riders* will need to tweak variables to suit themselves, but this won't change if for MTB vs track vs. road.

I see the formula being a function of CP values over a broad range, with stress levels calculated based on how close you came to hitting those numbers during your ride.
formerly rruff...

Rick
Posts: 2001
Joined: Sat Aug 29, 2009 4:30 pm
Maybe I am just not imaginative enough; but let's says that I got an equation that gives me a perfect "NP" measure, and it has been tailored with parameters to fit me perfectly. Since it is a "NP" that requires a calculation taking into account all the data, it can't be calculated or read until after the ride, or at least the after effort I am analyzing, is over.
So, what do I do with the number ? Just know that I went too hard or not hard enough ?

Maybe it is valuable in some sort of long-term trend analysis ? I don't know.

jmorgan
Posts: 21
Joined: Thu Jan 17, 2013 3:37 am
Seems like everyone wants something to tell them what they can or could do without ever actually doing it lol. I wonder if these same people use FTP or CP values from several months ago that are longer valid also lol.
2012 Cervelo S5 Team

jmorgan
Posts: 21
Joined: Thu Jan 17, 2013 3:37 am
WMW wrote:With a more sophisticated model, you'd be able to tweak some variables to customize it to yourself. It is going to be a moving target anyway because your CP numbers change... but NP is a lot cruder than it needs to be IMO.

Essentially this is what Andy Coggan has come out with for WKO 4.0 with his new model. The thing is he even says it only as good as the data that it is derived from. If all you ever do is 2x20min intervals its not going to be good at predicting your 2min power or 3 hour power. I'm sure it can figure something out but "the best predictor of performance is performance itself".

NP also has nothing to do with CP and it changing. As your CP increases your NP should also increase is about the only correlation between the 2.

NP is not a model of performance. NP is a weighted average. Stop trying to make NP more than what it is, its not a model and never will be. My guess is you used to look at NP and saw big numbers over short intervals that made you happy until you realized you couldn't sustain those numbers because NP is supposed to be used over a longer duration than your short intervals and now you hate it because you are not using it correctly.
2012 Cervelo S5 Team

Rick
Posts: 2001
Joined: Sat Aug 29, 2009 4:30 pm
Just letting my imagination run wild here:
But maybe, with a knowledge of your own current CP curve, Strava-like historical GPS and prior effort data, and a sophisticated enough algorythm for an "NP-like" parameter you could reach the point where, as you approach a hill the computer could warn you that even though you can usually carry 350 Watts all the way to the top, you better not go above 270 right now or you'll blow.

WMW
in the industry
Posts: 855
Joined: Sun Jul 29, 2012 2:59 pm
Location: Ruidoso, NM
jmorgan wrote:My guess is...

Really poor guess. My disappointment is that it models reality poorly. The same goes for TSS.

NP is used to translate a variable effort into a pseudo steady-state effort... which is then used to determine training stress. NP and TSS are fundamental for tracking your training with power.

I've found TSS to be a poor gauge of training stress for me. Mostly because it greatly understates the stress of true 100% efforts. There is a world of difference between a 95% and a 100% effort, that is missed by the TSS value.

The reason for using CP values *instead* of the standard NP formula is because it could more accurately represent the effect of variable efforts.
formerly rruff...

WMW
in the industry
Posts: 855
Joined: Sun Jul 29, 2012 2:59 pm
Location: Ruidoso, NM
Rick wrote:But maybe, with a knowledge of your own current CP curve, Strava-like historical GPS and prior effort data, and a sophisticated enough algorythm for an "NP-like" parameter you could reach the point where, as you approach a hill the computer could warn you that even though you can usually carry 350 Watts all the way to the top, you better not go above 270 right now or you'll blow.

I would never it use it for that... don't use one for pacing unless it is the beginning of a TT.

The PM is for field testing and tracking training load and performance. For that to work properly the NP and TSS values need to accurately represent what they are intended to represent.
formerly rruff...

jmorgan
Posts: 21
Joined: Thu Jan 17, 2013 3:37 am
WMW wrote:
jmorgan wrote:My guess is...

Really poor guess. My disappointment is that it models reality poorly. The same goes for TSS.

NP is used to translate a variable effort into a pseudo steady-state effort... which is then used to determine training stress. NP and TSS are fundamental for tracking your training with power.

I've found TSS to be a poor gauge of training stress for me. Mostly because it greatly understates the stress of true 100% efforts. There is a world of difference between a 95% and a 100% effort, that is missed by the TSS value.

The reason for using CP values *instead* of the standard NP formula is because it could more accurately represent the effect of variable efforts.

Again Bike Score was built around the same thinking by Phil Skiba.
http://www.physfarm.com/bikescore.pdf

Conclusion
"Although it remains to be seen whether the above changes represent a substantial (indeed
any) improvement over Dr. Coggan’s rather robust solution, we suspect that more formal
studies will demonstrate no signiﬁcant difference between these two different solutions."

It does sound like Phil is waiting to release some more stuff soon though so maybe you will have a better model.

You also say there is a world of difference between 95% and 100%. Mentally sure, physiologically probably not that much. Energy wise, 300w CP, 25kJ W'. Flat 1000w sprint effort you would use your W' up in 36 secs thats 100%. 95%w over CP would be 965w or for time what a second less or using only 23.7kJ. You are taxing all of your systems very similarly at that point.

If you are affected so differently why not come up with your own personal score and base it on RPE, watts, time, CP, etc.
2012 Cervelo S5 Team

• Similar Topics
Replies
Views
Last post