Jump to content


Photo
* * * * * 1 votes

race pace 2019


  • Please log in to reply
10 replies to this topic

#1 moreland

moreland
  • Member

  • 92 posts
  • Joined: July 17

Posted 15 May 2019 - 18:36

Hi, it's a little late, but here is the 2019 edition of the race pace analysis thread. There are similar threads for the 2017 season here and 2018 season here. The idea is, we take all lap times by all drivers in each race, apply a common statistical method called linear modelling, and produce a best guess for what each driver's outright pace in each race was.
 

We'll start off by explaining the method. The explanation that I gave last year seemed to work so I've virtually cut and paste it here, so if you've read that recently, this might sound familiar. I have however

more recent example to explain the method, I've chosen Romain Grosjean from Spain 2019 since he had a fairly typical race, with many laps in clear air but also a few in traffic.

So, as I said, we want to take the lap times for each driver in each race and from them, calculate what their outright pace was. Each lap time is affected by the amount of fuel on board, the tyres used and the age of the tyre, so we need to estimate the effects of those. In addition, lap times can frequently be affected by being stuck behind other cars, or being involved in overtaking battles, so we filter out these laps.

So let's take a look at Grosjean's race in the 2019 Spanish Grand Prix. Here are all of his lap times:

rgrosjean-raw.png

 

This plot is not great because the scale is dominated by the very slow laps around the pit stops and safety car period, so we'll remove these unrepresentative laps. In general we remove laps done behind the safety car, inlaps, outlaps, lap 1, safety car restart laps and obvious outliers. Next, we'll plot the remaining lap times, but we'll highlight when Grosjean's lap times were affected by traffic. These were at the start of the race before Gasly pulled away, and after the safety car restart while he was involved in battles with Magnussen, Sainz and Kvyat. In general I remove any laps where they overtook someone, got overtaken, got lapped or were blocked. Having done various investigations, I use a definition of being within 1.5 seconds of the car ahead as being blocked.
 

rgrosjean-blockedot.png

 

So the 'clear lap's are the useful ones. However, there are factors that affect these lap times, most importantly the fuel load, which tyre was being used and how long it had been used for. So we'll start off by displaying the effect of the fuel load. Below is the plot of Grosjean's 'clear lap' times, but with a line of best fit going through them - this has been obtained by finding the line that passes most closely through all of the lap times for the clear laps. It doesn't pass very closely between them, but we'll deal with that shortly.
 

rgrosjean-fuelfitted.png

 

Using that line, we can display the lap times for Grosjean correcting for the fuel load. These can be thought of as the lap times he would have set if the car had had only 1 lap of fuel in it, and are displayed with the green points here:
 

rgrosjean-fuelcorrected.png

 

So the biggest trend i.e. lap times dropping throughout the race, has been accounted for. However we notice that the lap times get slower as he gets further into each stint. This is of course the effect of tyre wear. The tyre wear is generally more difficult to estimate accurately than the fuel load because there is quite a lot of variation in how drivers treat their tyres in the race, there are phases when they push and phases when they drop back etc.

There is also the question of whether the tyre wear is different for different cars. There might well be some variation but in this model we group all cars together when estimating the effect of tyres (and the fuel load for that matter). By doing this we get estimates that are certainly close to being accurate for every car and even in the cases where a car does use their tyres a little differently to others, it's rare that it has much of an impact on their final pace estimate because the effects of it tend to balance out over the course of the race.

We'll discuss the tyre wear in more detail later on, but for now we'll display what the effect of the fuel and tyre was for every lap. However, when we say 'tyre effect', it's a bit ambiguous, the effect compared to what? We need to choose a 'reference state' and display how much time was lost compared to it. When we referred to the fuel effect, the reference state was 1 lap of fuel, i.e. we calculated what lap time would have been set with 1 lap of fuel. For tyre, the chosen reference state is a new set of the fastest tyre (when new). As it turns out, the soft tyre is calculated to be the fastest tyre when new.

So here is the display of how each of Grosjean's lap times were affected by fuel, tyre choice and tyre wear on every lap:
 

rgrosjean-fuel-Tyre-Effect.png

So we see that we take off less and less throughout the race, as his fuel load goes down. However at the end of the race we still take off almost a second: although he's only got 1 lap of fuel left, his softs have worn down so that's costing him about a second compared to a fresh set.

As before, we then take these numbers away from his actual times to get fuel and tyre corrected times:
 

rgrosjean.png

 

The green line represents the average of all of Grosjean's lap times, accounting for his fuel load and tyres. If Grosjean had driven exactly the same way on every lap, and model was completely correct in estimating the way the Haas used its tyres and fuel loads, then all the green points would lie exactly on the green line. Obviously that isn't the case, so we see a certain amount of scatter.

Grosjean's scatter doesn't have too much of a trend to it, but if we look at the equivalent plot for Hamilton for example, we see that he hasn't set lap times quite in accordance with the calculated adjustments for fuel and tyres.
 

lhamilton.png

 

He's faster than we expect in stint 1 but slower in stint 2. This could well be explained by Hamilton's need to push in the first stint in order to build a gap to Bottas prior to his first tyre stop, but to then look after the tyres to the end of the race (which turned out not to be necessary due to the safety car) in the second stint, which he could afford to do given he had an 8 second lead over Bottas during that phase of the race.

So we repeat this process for all drivers and we get the 'average of fuel and tyre corrected lap times' line for all of them, and these represent their race pace estimates. Here are how all the drivers are ranked in the Spanish Grand Prix:
 

2019spain-Race-Pace.png

 

So Hamilton is rated as the fastest driver, unsurprisingly since he won the race quite comfortably. As mentioned above, he could probably have pushed a bit harder in the second stint if necessary so his advantage possibly could have been bigger if he'd needed it to be.

Vettel is rated a little faster than Leclerc, this is a little hard to judge since they were on different strategies but Vettel is rated a little faster mainly due to some very quick lap times in his second stint. One little quirk is that Leclerc's stint in the hard tyres don't count towards his rating, because the hard tyre was not used by many drivers thus we can't estimate what the effect was on lap times. For the same reason, Giovinazzi does not have a rating, since almost all of his clear lap times were on the hard tyre.

 

It should be pointed out that the estimates are subject to quite a bit of uncertainty. For example, Norris is rated a little faster than Sainz, but if we compare their lap times side by side, we can see that both of their races were heavily disturbed by traffic and that Norris's advantage is by no means consistent throughout the race:

 

csainz-lnorris.png

Meanwhile, here are the graphs for the rankings for the other races this season. I'd say that all races have been quite suitable for the analysis because they've all featured a reasonable amount of field spread. I think it's fair to say that the races have overall been somewhat uneventful so far, but on the bright side, the less chaotic the races, the more reliable these rankings are! If you've got any queries about certain race/driver estimates, I'm happy to display the plots detailing how the calculations have been done.

 

2019australia-Race-Pace.png

 

2019bahrain-Race-Pace.png

 

2019china-Race-Pace.png

 

2019azerbaijan-Race-Pace.png

 

Also, here are the team mate comparisons:

 

mercedes-2019-30.png

 

ferrari-2019-30.png

 

redbull-2019-30.png
 

mclaren-2019-30.png

 

renault-2019-30.png

 

tororosso-2019-30.png

 

alfaromeo-2019-30.png

 

haas-2019-30.png

 

racingpoint-2019-30.png

 

williams-2019-30.png

 

Here is the overall average for the season so far:
 

2019-yearly-pace.png

 

Don't think there can be too many arguments with which team heads the field. The big surprise has been Haas's slow race pace, although they did recover nicely in Spain. Gasly seems to be destined to have a lot of very lonely races this season, while Kvyat is currently the leader of 'Formula 1.5' but I expect the order of that battle will change frequently throughout the season given how close it is. Albon's average is a bit dubious, it's dragged down a lot by his race in Australia, which was a lot slower than Kvyat but is not representative of his pace in general.

As before, all comments and questions are welcome, although I might not be able to reply quite as quickly this year as in the past. I probably won't be able to update after every race this year for that matter, but everything should appear at some point throughout the year :)


Edited by moreland, 15 May 2019 - 19:18.


Advertisement

#2 f1paul

f1paul
  • Member

  • 6,623 posts
  • Joined: April 16

Posted 15 May 2019 - 19:01

Glad you can still do these Moreland, even if it's only every few races.  :up:



#3 FullOppositeLock

FullOppositeLock
  • Member

  • 4,849 posts
  • Joined: September 15

Posted 15 May 2019 - 19:49

Excellent work moreland, and many thanks for all the effort put into this analysis.

Your model must have a flaw somewhere though because for some odd reason it’s not showing that Renault’s race pace is much quicker than other midfield cars (just kidding in reference to a lenghty discussion earlier today in the Renault car thread ;))

#4 Requiem84

Requiem84
  • Member

  • 5,164 posts
  • Joined: September 10

Posted 15 May 2019 - 20:20

Great stuff! Very much appreciated!!

#5 sgtkate

sgtkate
  • Member

  • 262 posts
  • Joined: May 17

Posted 15 May 2019 - 21:53

This is really interesting. Thank you.
Funny how Azerbaijan seems to be an outlier circuit. A lot of 'underdog' team mates beat the 'faster' team mate there, 4 teams. I wonder why?

#6 coppilcus

coppilcus
  • Member

  • 1,454 posts
  • Joined: November 12

Posted 16 May 2019 - 07:02

Thank you!

#7 Maxioos

Maxioos
  • Member

  • 3,201 posts
  • Joined: October 17

Posted 16 May 2019 - 07:29

Thanks.

 

I struggle to understand why Vettel would have had better race pace than Max last GP as in Bahrein where he finished 30 seconds behind Max. If you have a pitstop extra, 2 stints on C3 against 2 stints on C2, it's not a surprise you have higher race pace seems to me.



#8 sgtkate

sgtkate
  • Member

  • 262 posts
  • Joined: May 17

Posted 16 May 2019 - 10:01

Thanks.

 

I struggle to understand why Vettel would have had better race pace than Max last GP as in Bahrein where he finished 30 seconds behind Max. If you have a pitstop extra, 2 stints on C3 against 2 stints on C2, it's not a surprise you have higher race pace seems to me.

The aim of the stats and calculations is to try to remove that affect on the data. In theory it shouldn't matter if a driver took 4 pits stop or 1 or had different tyres as this should get normalised out. I can imagine that is INCREDIBLY hard to do though.



#9 sopa

sopa
  • Member

  • 11,681 posts
  • Joined: April 07

Posted 16 May 2019 - 10:19

Kvyat has on average the 7th best race pace this season? A shocker.



#10 Otaku

Otaku
  • Member

  • 1,339 posts
  • Joined: March 11

Posted 16 May 2019 - 16:24

Nice to see Ferrari and Red Bull so close. Kvyat is impressive, they should switch him with Gasly.



#11 moreland

moreland
  • Member

  • 92 posts
  • Joined: July 17

Posted 17 May 2019 - 08:05

Excellent work moreland, and many thanks for all the effort put into this analysis.

Your model must have a flaw somewhere though because for some odd reason it’s not showing that Renault’s race pace is much quicker than other midfield cars (just kidding in reference to a lenghty discussion earlier today in the Renault car thread ;))

 

Thanks! To be fair, Renault have been hard to assess this year, their pace seems to be quite up and down and they're struggling to put clean weekends together, a bit like Toro Rosso for much of last year.

 

This is really interesting. Thank you.
Funny how Azerbaijan seems to be an outlier circuit. A lot of 'underdog' team mates beat the 'faster' team mate there, 4 teams. I wonder why?

 

I noticed that too, the formbook went out of the window generally there, with eg Perez being so quick and Hulkenberg being so far back. One theory I have is that there was only one tyre that was at all competitive in Baku (which was the C3/soft) and it had low deg, so how fast a team was came down to how well they could make that tyre work for the entire race.

 

Thanks.

 

I struggle to understand why Vettel would have had better race pace than Max last GP as in Bahrein where he finished 30 seconds behind Max. If you have a pitstop extra, 2 stints on C3 against 2 stints on C2, it's not a surprise you have higher race pace seems to me.

 

In Bahrain Vettel lost a lot of time with a spin which required a pitstop for a new nosecone so lost a huge amount of time there. Here are Verstappen and Vettel's lap times in Spain:

 

svettel-mverstappen.png

 

It's the second stint that drags Verstappen down a little, to be fair he's comfortable at this stage of the race and has no need to go any faster. But when you factor in his tyre and fuel his pace was about 0.5 seconds/lap slower than his pace in stints 1 and 3.

 

The aim of the stats and calculations is to try to remove that affect on the data. In theory it shouldn't matter if a driver took 4 pits stop or 1 or had different tyres as this should get normalised out. I can imagine that is INCREDIBLY hard to do though.

 

Yes, you're right, the model tries to factor in what the effect of tyre strategy was and adjust for it when producing the outright pace estimates. With the different strategies, and drivers having phases of pushing and easing off, being stuck behind other drivers etc, I agree the reality is more complicated than what my model adjusts for. So in some situations the estimate will be a bit out for certain drivers in certain races but over the course of the season it should balance out. I would argue it is much better in this respect than the points table, for example, in 2018, Ricciardo retired from 8 races. For those 8 races, the points table awarded Ricciardo the same score as the slowest driver in the field. My rankings never do anything as wrong as that 8 times to the same driver in a single season!

 

Kvyat has on average the 7th best race pace this season? A shocker.

 

That's all about consistency, the midfield is tight and his slowest performances haven't been too slow, unlike most of his rivals.

 

Looking at the estimates for each race you might expect Sainz to be a bit higher in the overall average and it's worth explaining why he isn't. He doesn't have an estimate for the Australian GP and has been awarded a 'phantom' slow rating for it. The reason is, I don't want the average to be affected by which races a driver did or didn't get an estimate for, so when a driver doesn't get an estimate for a race, I make a guess about what their estimate would have been, then include it in the average. Norris has been slightly faster on average, so my best guess for how fast Sainz would have been in Australia is that he'd have been slightly slower than Norris. This then drags down Sainz's estimate (and likewise, Norris gets a phantom fast rating for the Chinese GP, which he didn't get an estimate for, and this helps his overall average). That might sound a bit weird and complicated, but I think it needs to be done, otherwise a driver can be flattered compared to their team mate by missing the races where their car wasn't competitive. This happened in 2017 when Button only did the Monaco GP, which was clearly Mclaren's best track, therefore (before doing this fix) he was rated as the fastest Mclaren driver in 2017 which made no sense (especially given that Vandoorne was in fact rated as faster than Button at Monaco). Hope that all makes sense... :stoned:


Edited by moreland, 17 May 2019 - 08:07.