In my last post I argued that using the average heart rate of a ride as measured by my TranyaGo heart rate monitor to calculate Load would be a better way to estimate how much riding I have been doing than using just the length of the ride in minutes (which is what I had been doing). In the month since that post, my experience continues to support that conclusion. A few weeks ago, when I had just started ramping back up from a lull in my training caused by bad weather, I was feeling very tired towards the end of the week at a point where my schedule called for one more ride. So, listening to my body suggested skipping that ride. However, communication from my body is not always reliable and my training plan seemed reasonable; normally, I try to ride between 300 and 400 minutes a week and I was at only 257 minutes. However, when I looked at my Load for that week, it was actually quite high due to the fact that some of the rides that week had been ridden fairly fast. Putting that all together, I decided to skip the ride, a decision I would not have reached without the Load data. Of course it is impossible to know with certainty if that was the right decision, but my intuition tells me that it was. In this post, I am going to discuss extending that one more step: would it be an additional improvement to explicitly track the accumulation of Fatigue over time using the model of Banister?
Banister’s model is a tool for using Load data to predict Fitness, Fatigue, and Form. I blogged about this model back in 2021. It is similar to the TrainingPeaks software used by many cyclists. Unlike TrainingPeaks, the Banister model is not available as a package, I implemented from the description in Banisters paper in a spreadsheet. I could have just purchased the TrainingPeaks software but I chose to implement the Banister model instead because I understand it, it is fully customizable, it was designed to be used with heart rate data whereas the TrainingPeaks software was initially designed to use Power data from a power meter, and because the Banister model is free. The figure at the top of this post illustrates the application of Banister’s model to my average heart-rate derived Load data that I have collected since I purchased my TranyaGo.
What are Load, Fitness, Fatigue, and Form? Load is how hard a ride is, which is a combination of both how intensely I ride (e.g. how fast, how hilly, etc.) and how long I ride (e.g. how many minutes.) Fatigue is how tired I am as a result of all the rides I have been doing. Obviously, a ride I did yesterday has more impact on today’s Fatigue than a ride I did six weeks ago, but they both have some impact. Fitness is kind of a hidden quantity. It is how strong I would be if I had no Fatigue. Form is how strong I actually am given both my Fitness and my Fatigue: Form = Fitness - Fatigue.
Besides assuming that the overall model is correct, Banister’s model as published assumes the values for four parameters; how much a given amount of Load adds to Fitness, how much a given amount of Load adds to Fatigue, how quickly an athlete recovers from Fatigue, and how quickly an athlete loses Fitness. In that original publication, Banister assumes that both Fitness and Fatigue decrease exponentially over time and that Fatigue is reduced by half after about 10 days and that Fitness is reduced by half after about 30 days. It also assumes that one unit of Load initially increases Fatigue twice as much as it increases Fitness. The combination of all these assumptions is that initially, a ride reduces an athlete’s Form (e.g. they cannot ride as fast or as long) but over time, Fatigue decreases faster than Fitness so that an athlete’s Form will increase to a new, higher level. This is exactly what is predicted by virtually every coach and exercise scientist, this is arguably the central dogma of exercise. What is open to debate, however, is how fast and by how much, things that are determined by the model and the values of the parameters used by the model. The figure at the top of this post uses the values of these parameters initially published by Banister.
Way back in 2018, years before I blogged about the Banister model, I blogged about a paper, Busso et al., that I now realize was based on the Banister model. It extended that model in that it considered the four parameters discussed above not as fixed, but as variable depending on the exercise program. I am skeptical about that latter claim but I was interested in the general ranges for these variables they came up with. More or less arbitrarily I picked values within these ranges, values that were different than those assumed by Banister. The parameters for Fitness did not change much, but for Fatigue, the Busso et al. value for impact of Load on Fatigue was 1.3 times that for Fitness as opposed by the value of 2 proposed by Banister, and the time for Fatigue to decrease to half was 7 days as opposed to the 10 days proposed by Banister. I replotted my data with these new values as is shown below:
Changing the parameters made a big difference. I am not so concerned with the differences in the absolute values of Fitness, Form and Fatigue, I don’t think those are meaningful, I am more interested in the differences in the shapes of the curves, in particular, that for Form for the past two months. During the first half of this period, my cycling was interrupted by some extremely bad weather and the result is that Fitness, Form, and Fatigue all fell. This is true both when I use the Banister parameters and when I use the Busso et al. parameters. During the second half, the weather improved and I was able to increase my cycling back to what I had been doing. This increased both Fitness and Fatigue as expected but depending on how much it affects each of these, Form, my actual ability to ride, could increase, decrease, or stay the same. When I use the Banister parameters, it decreases. When I use the Busso et al. parameters, it increases. My somewhat subjective observation of my cycling ability is in between those two, I would say it is staying about the same with maybe a slight increase overall. These differences matter. If my Form is increasing, it means I can increase the amount of training I do by riding longer and/or faster. If it is decreasing it means I am training too much and should cut back. If it is staying about the same, then I am doing about as much training as I can sustainably manage and should continue at this level of training for now. So what should I do?
When I started this experiment, I did so thinking it would be a way to use Fatigue to track the effect of Load over time by allowing me to consider the effect of rides done before the current week while including the fact that the longer ago a ride was, the less impact it would have today. I thought that tracking Fitness and Form would not be useful. When I looked at the output, I realized that Form was a much better indicator of what I was interested in than Fatigue, that Fitness was also of some value, but that Fatigue, what I had assumed would be the most useful, turned out to be of no apparent value in and of itself. (Obviously, it is an essential part of the model, necessary for the calculation of Form.) I also realised that a graphical view of Form and Fitness were much more useful than a numeric ones. Because I had the good fortune to have come across two different papers using the Banister model, the original one by Banister where the model was introduced and a later one by Busso et al. that used different values for the parameters, I realized how important tuning the values of these parameters to me specifically will be. What I plan to do is to try different values of parameters until I get an output that matches how I feel. This will not be exact because how I feel is both subjective and dependent on things other than my training (illness, emotional stress, and non-cycling activities, for example.) So why then not just rely on how I feel? If I can tune the Banister model to my current physiology, it will give me a systematic, objective estimate of where I am in my training cycle, an estimate I can use along with my subjective feelings and my pre-existing training plan to help me decide whether to skip a planned ride, substitute an easier ride, ride as planned, or do more than I planned thus investing in my store of Fitness.
