Hi, I’m currently on a trial and am curious how AIE determines FTP. I’ve only just started recording DFA A1, so there is not enough quality data there. In this situation the LLM tells me its determining FTP base on a Critical Power Model/Curve based on my training history, similar to Intervals.icu. Is this correct? I can’t find any reference to this approach any where.
Out of curiosity, I looked as it used to be in the ‘Account’ section as ‘FTP for zwift’, but I don’t see it any longer. Maybe Markus can elaborate on why it is no longer there.
However, in my current training plan which I asked to optimized critical power I can see it in the ‘Data’ page (historical) and on the Dashboard (predicted increase).
What are you interested in using the FTP for? AI Endurance doesn’t use FTP when planning workouts, as I understand at least, unlike TrainerRoad or Zwift (for example).
Hi. Thanks for the response. Yes I can see my calculated FTP in the account section. My query is the more about how that value to calculated so that I can better understand how AIE works.
I have AIE set up with a goal of improving Endurance Power. The fact that AIE determines your LT1/VT1 threshold independantly of FTP is one of the attractions, however the FTP value it calculates is still important when undertaking Threshold or VO2max workouts. Hence my question - when there is no DFA A1 data or test data, is FTP derivied using a Power Model/Curve from available training data similar to Critical Power? If yes, what are the key interval duration for deriving that curve?
As I enjoy thinking about this kind of thing, I’ll recount my understanding, but don’t take my word for it (I don’t work for the company after all, this is just my understanding based on Markus answer he gave in this thread: How are the training zones used during plan generation?
I suppose that’s the difference. The mapping of ‘threshold’ or ‘vo2max’ is, in my understanding, a labeling convention and not used when generating the workouts. The ML model determines the optimal pace/power for time within constraints (sports being trained, availability per week, date of event, type of event, etc.) and then maps this onto workouts which are given the label ‘threshold’ or ‘vo2max’.
Put another way, the watts to produce for a certain duration comes first, and then sliced into a 14-day microcycle where individual workouts are created, and within that 2 week period if you need X minutes at A watts, you get a suitable number of workouts where you, if done with 100% optimal adherence, hold the A watts for X minutes over the course of that microcycle. When creating the individual workouts it groups ‘like’ wattages together into various labeled buckets (endurance, tempo, vo2max, anaerobic, etc.) which show up on your training calendar. This explains, in part, how you might have a ‘threshold’ workout with one 9 minute threshold interval and 30 minutes of endurance.
This imprecision sounds a bit weird (what are the cutoffs between endurance/temp, is it halfway through? Ceiling/floor functions? Rounded up/down? etc.), and then I recalled Xert has the magic buckets which let you overshoot intensity for a given ‘bucket’ and it compensates by reducing the duration of the interval, and that’s a pretty mathematically driven organization (but they don’t do run training so … not for me), and TR does something similar with their ‘daily workout’ suggestions (but not season planning, at least not according to the most recent episode with CEO Nate Pearson).
None of that explains how Endurance Power, Functional Threshold Power or Critical Power is calculated. Though from my understanding, FTP has no physiology-based definition (assuming you take Andy Coggan’s word on how the idea behind FTP was to not require learning the physiology, which he elaborates on in an episode of the Inside Exercise podcast https://youtu.be/-6DhfMJH84E?t=1760 (timestamp jumps straight to this discussion). As “evidence” on the non-standardization of FTP, TrainerRoad has redefined FTP in their platform to a ‘level 3 threshold’ (whatever that is), but whatever it is was not in “Training and Racing with a Power Meter”.
I hope Markus can answer your questions, I am also interested in the answers.
@aindriu_b it’s a combination of:
-
Critical Power via power duration curve (3 - 12 min durations)
-
DFA a1 thresholds (if available)
-
20 min power records
-
auto detection of tests (ramp, 20 min mostly)
Without DFA a1 data, power duration curve is what’s most relevant when riding without dedicated testing workouts.
We use heart rate during the efforts to gauge ‘how hard you went’ and give each measurement a relative weighting (you can see it in the My Past Data chart in the Data page). The higher the weighting the more we trust this was an effort measuring your true potential at any given time.
Because Zwift lets us define direct power targets in the new direct integration. No need to sync threshold settings in both apps anymore for Zwift. Still necessary for some though (Rouvy, TrainingPeaks, TrainerDay)
Thanks for the detailed summary. Yes the relationship between FTP, CP, LT2, VT2 is a bit murky considering they’re all informing the ‘anaerobic threshold’. Probably the most sensible thing is to define each metric by what it actually measures. For example,
- FTP is 95% of best 20 min power
- CP is the fit parameter of fitting the CP model towards the power duration curve
- and so on…
In practice getting high quality measurement on any of them and aggregating all you can get seems a good approach. Coverage over measuring your true performance potential is generally speaking a bigger limiter than the difference between the relative measures in my experience.