Water meter with ai on the edge

Energy
,
1

Hi all,

I have an esp cam with AI on the edge installed reading my analog water meter. Two days before I notice a glitch in the dashboard.
Here are my findings:
in SQLite - querying the internal HA DB - I found this:

SELECT 
datetime(created_ts,'unixepoch') as created,
datetime(start_ts,'unixepoch') as start,
state, sum
 FROM "statistics"
where metadata_id = 194
order by created_ts desc

resulting in this output (metadata_id is sensor.watermeter.value, state value is in qubic meters):

created 		        start 			        state 		sum
2025-06-13 18:00:10 	2025-06-13 17:00:00 	1001.158 	4608.343
2025-06-13 17:00:10 	2025-06-13 16:00:00 	1001.156 	4608.340999999999
2025-06-13 16:00:10 	2025-06-13 15:00:00 	1001.144 	4608.329
2025-06-13 15:00:10 	2025-06-13 14:00:00 	1001.136 	4608.321
2025-06-13 06:00:10 	2025-06-13 05:00:00 	999.99 		93.53100000000006
2025-06-13 05:00:10 	2025-06-13 04:00:00 	999.99 		93.53100000000006
2025-06-13 04:00:10 	2025-06-13 03:00:00 	999.99 		93.53100000000006
2025-06-13 03:00:10 	2025-06-13 02:00:00 	999.99 		93.53100000000006
2025-06-13 02:00:10 	2025-06-13 01:00:00 	999.99 		93.53100000000006
2025-06-13 01:00:10 	2025-06-13 00:00:00 	999.99 		93.53100000000006
2025-06-13 00:00:10 	2025-06-12 23:00:00 	999.99 		93.53100000000006
2025-06-12 23:00:10 	2025-06-12 22:00:00 	999.99 		93.53100000000006

Once the state flipped from 3 to 4 digits it seems to have a glitch in building the sum, resulting in this output on the enery dashboard:

grafik
grafik1283×356 19 KB

Is there a chance of a bug? Did I do something wrong in the configuration?
Thanks for any hint.

Best,
MikeE

2

Can you get a dump of that entity’s recorder history from 05:00:00 to 15:00:00?

Is there a reason why there’s a big time gap there?

3

Hi,
the gap between happened because the water meter did not recognise the numbers correctly. There were some “negative reads” on the ESP, this is why the gap shows.
Thanks,
Mike

4

That’s probably where your error is coming from. You’ll need to sanitize the data a bit before the statistics gets hold of it or it will generate bogus results.