C'est une forme de vides et d'îlots. Vous pouvez utiliser une différence de numéros de lignes pour obtenir les îles :
select device_id, speed, count(*) as num_times
from (select t.*,
row_number() over (partition by device_id order by datetime) as seqnum,
row_number() over (partition by device_id, speed order by datetime) as seqnum_s
from t
) t
group by device_id, speed, (seqnum - seqnum_s);
Ensuite, pour en tirer le maximum, utilisez une autre couche de fonctions de fenêtre :
select device_id, speed, num_times
from (select device_id, speed, count(*) as num_times,
row_number() over (partition by device_id order by count(*) desc) as seqnum
from (select t.*,
row_number() over (partition by device_id order by datetime) as seqnum,
row_number() over (partition by device_id, speed order by datetime) as seqnum_s
from t
) t
group by device_id, speed, (seqnum - seqnum_s)
) ds
where seqnum = 1;