Sense forgot my EV (Tesla)

technicalsupport
#1

One of the key reasons I bought my Sense was to track my Electric Vehicle usage.

After about 10 weeks, Hooray, Sense found my Tesla Model S (which I learned here is a device Sense claims to know all about). I charge my Tesla very nearly every night at midnight.

But Today, Sense failed to recognize my Tesla charging at midnight, despite seeing its vivid signature on the “Now” timeline.

For some reason (beyond me), Sense went back to learning my EV (as reported - My Home: Device Detection screen).

Anyone else experience this? Ben@Sense, any ideas??

:frowning: Ed

2 Likes
#2

Hey Ed,

Sorry to hear that Sense missed a charge there. Has Sense since picked up other charges? Our Tesla Model S detection algorithm still has some room for improvement (and might occasionally miss a charge), but it should detect most of your charges. Totally understand the frustration with this given the high wattage of the device, and it is something we’re working on.

With regards to it being reported in that learning section, that section is in need of some improvements itself for clarity and accuracy. In the case of your Model S, it likely appeared there because Sense is refining the device model. This refinement is a normal part of the detection process as Sense sees more instances of the device running in your home.

1 Like
#3

Good News Ben,

My Sense only missed one instance Telsa charging, Saturday a.m. starting at 12:00 a.m. and lasting into the 7:00 a.m. hour.

When I realized Sense had missed this detection, I decided (in the Sat. 11:00 a.m. hour) to add some more charge to the Tesla.

Sense picked up this charge. And then it again recognized my next scheduled charge on Sunday a.m. at midnight.

So it appears to be detecting the Tesla properly again.

BTW, I tried to delete my Blog entry on Sunday, but couldn’t do it. I will update it with this info instead.

Thanks for your quick, thoughtful and descriptive response.

Ed

3 Likes
#4

Glad to hear it!

2 Likes
#5

Ben, I have a Tesla S as well using their Wall Charger @ 48A. Its at a winter vacation house so we don’t charge there ever day (obviously). Its also on a TOU plan so charging starts after 8:00pm for us. Any way to “kickstart” the discovery process? Thanks

#6

@svollum, unfortunately there isn’t much you can do to kickstart it. It really just comes down to Sense seeing it charging there a number of times, until it is confident that it has detected the signature of a Model S. You won’t need to charge it there daily, but it will likely take a couple dozen charges for it to be detected.

1 Like
#7

Hi Ben, Sense has again forgotten my Tesla Model S. I originally posted(11 days ago) that it missed a couple of charges, but then re-started to recognize charging which lasted for a couple days. All was well for a while.

Now it has again stopped recognizing charging… the last 5 or so charges. Sense went back to learning my EV (as reported - My Home: Device Detection screen)

:frowning: Ed

#8

Hi @Ed_Texas - very sorry you’re encountering this issue.

We’ve logged a bug with the data science team to try and root this out. Their initial thought when I sent it over was that this could be a complex issue that will take some time to figure out, so we may not have an immediate solution for you, but please let us know if the status changes.

Sorry again for the frustrations here. Thanks for your patience!

1 Like
#9

Sense has again forgotten my Tesla Model S twice now. I originally posted(38 days ago, then again 27 days ago) that it missed a couple of charges, but then it re-started to recognize charging which lasted for a couple days. All was well for a while.

Then it again stopped recognizing charging… the last 5 or so charges. Sense went back to learning my EV (as reported - My Home: Device Detection screen).

After supposedly learning my EV again, despite registering the EV Device as an official device, it failed to recognize all “charge” cycles. After about a week of of not recognizing any charges, I decided to manually delete the device.

That was nearly a month ago - No EV device!

Incidently, my SENSE hasn’t recognized ANY NEW DEVICES in this past month… hmmmm…
I there something else malfunctioning?

Ed

[/quote]

#10

Ed,
If it’s any consolation, our Model S charging detection (80A) via Sense has disappeared a few times, but eventually come back. It’s been working for us reliably for about 6 months. My only regret is that they still can’t detect my 72A Model X charge cycle that consumes another 20% of the total usage hidden in “Other”.

1 Like
#11

Hi @Ed_Texas - very sorry this has been so frustrating for you. The team does have a bug about this specific issue and is working on it. Unfortunately, it’s fairly complex so they’re still working through it. We’ll be sure to update this thread when we know more!

Sorry again for the disappointment here, and thank you for your patience as we get it figured out. With this broader issue under control, we’ll be able to do a better job with detection across EVs, and hopefully start addressing vehicles like @kevin1’s Model X that still haven’t been detected.

Best,
Brad

1 Like
#12

Thanks…this is not the first time I’ve seen this behavior, and at one point (perhaps 6 months ago), I believe it was said to be a bug and to delete “discovered devices with zero data”. I’ve kept the three recent ones in case someone wants to take a look.

One of the three recent examples…the others were icemaker and light

#13

For the 4th time in as many months, my Sense has again found my EV (Tesla Model S).
Thereafter, however, it doesn’t recognize any usage (charging).

Strange. It surprises me that Sense can’t consistently track this EV. The Tesla charger has a very distinct current ramp-up in the first 30 seconds of operation “signature”, besides having the highest energy consumption in our home.

#14

I have installed 2 Sence monitors specifically to track my Tesla charging activity on the 2 circuit panels in my home. I have a 2017 Tesla Model X P100D that I charge 2 different ways, depending on where it is parked:

Tesla Mobile Charger - NEMA 14-50 @ 50amps
Tesla High Power Wall Charger @ 100amps

Any suggestions to accelerate the learning process or improve the probability of detection?

#15

As you can see from my post back in February, Sense eventually learned my Model S which uses a Tesla Mobile Charger - NEMA 14-50 @ 50amps. But then after about a week, it forgot it. I inquired about “why” it forgot it. The Sense people told me they have a bug…

Then a couple week later, it re-learned the Tesla charger. But that only lasted for a few days.

Bottom line, it recognizes the Tesla for days or a week or two. Then it forgets and re-learns after a couple of weeks.

Honestly, I have given up. The Tesla charger has a very distinct signature (I can say as I am a EE).
Can’t figure why they can’t make it work consistently over time.

Also, Sense’.s current policy doesn’t allow the user to speed up the learning process. There are theoreticians/purists within the company that insist the appliance learning process MUST follow the theoretical algorithms.

I have given up… :frowning:

#16

@Ed_Texas,

Read @RyanAtSense 's reply at the end of this thread. An EE should understand his explanation why Sense is somewhat flakey on EV detection right now, and why their new work on EVs should help. I can’t complain horribly since I have one Tesla, a Model S on the HPWC that is detected close to 100% reliably (but an X and a 3 with zero detection )

1 Like
#17

It’s not just a matter of the signature looking distinct, as I wrote in the post that @kevin1 linked to. The signature definitely does look distinct, it’s just a matter of getting the ML algorithm to reliably find it.

And it’s not just policy that disallows users from speeding up the ML process. We would certainly do that if it was feasible…there’s no purism here that wants us to keep our customers (and us…a lot of use here use Sense as well and don’t have 100% device detection) dissatisfied. We’re working on better ways to gather ground truth data. NDI was a start, but we’re looking into other possibilities. It’s a slow process, but we’re deep into research and experimentation on that front.

1 Like
#18

I understand, but I (and apparently many others) purchased Sense and put hundreds of hours into it; all based on Sense marketing (and the TOH hype). What I wanted/expected was for Sense to be able to detect and track my major consumers, and it mostly hasn’t.

Nowhere in the process was there disclosure that many cases weren’t solved, and weren’t going to be for a LONG LONG time…if ever.

Personally, I never expected, or needed 100%, but the best Sense has been able to do for me is under 10% of my devices, and that not reliably.

Andy

#19

All

Just to add to the frustration. I have 2 Teslas X and 3 and sense never picks up any of them. I charge one car at the time most days of the week…
I like me sense and it is great tool for understating the energy consumption but the ML is very disappointing as it can not pick up these obvious profiles. I have similar issues with my AC units too.

#20

I understand that the algorithm might need to look at very quick transitions to find some devices because otherwise they might look very much like another device, however when we are talking EV’s we are talking about huge draws over long periods of time, so perhaps the algorithm needs a fork that looks not just at quick transitions, but also for large loads that run for a long time. Perhaps if the algorithm is able to change the resolution that it is using it would detect some of these other loads. Think about it like this, if it missed the ramp up and ramp down, but captured the hours it is running at peak current, it would capture most of a typical home charge on an EV. It doesn’t matter if it recognizes that its a tesla or a golf or a leaf, but it should recognize this unknown large signal.

The first image is a VW eGolf, the next three are the piece parts of a Model 3 Charge, both are signals easily recognized on a large scale, but perhaps harder if the algorithm is focussed on a millisecond scale.

I gotta imagine the team has thought about this and looked at this, but it is super frustrating to have 68% of your load unknown despite the fact that the loads are easily recognizable by a human. I guess I am looking for a breakthrough here by prodding, cajoling, suggesting…