CODEWITHSUNDEEP
sqlpostgresqlgaps-and-islands
yglodt
yglodt

Reputation: 14551

SQL to find values higher than a threshold for a certain time

Consider the following table:

create table measurement (
    datetime     timestamp,
    temperature  numeric(5,2)
);

I would like to create an SQL-query in PostgreSQL which extracts the rows where the temperature was higher than 50 °C for at least 30 minutes, ideally knowing from when until when the temperature was actually above 50 °C. The example data is here:

datetime               temperature
-------------------    -----------
2017-03-15 19:00:10    49.56
2017-03-15 19:15:10    52.81
2017-03-15 19:30:10    49.00
2017-03-15 19:45:10    52.88
2017-03-15 20:00:10    49.56
2017-03-15 20:15:10    49.13
2017-03-15 20:30:10    51.31   <--
2017-03-15 20:45:10    52.06   <--
2017-03-15 21:00:10    50.50   <--
2017-03-15 21:15:10    50.50   <--
2017-03-15 21:30:10    49.38
2017-03-15 21:45:10    47.44
2017-03-15 22:00:10    46.19
2017-03-15 22:15:10    45.44
2017-03-15 22:30:10    50.25
2017-03-15 22:45:10    48.56
2017-03-15 23:00:10    51.25   <--
2017-03-15 23:15:10    50.44   <--
2017-03-15 23:30:10    50.63   <--
2017-03-15 23:45:10    46.75

Upvotes: 1

Views: 2066

Answers (2)

dnoeth
dnoeth

Reputation: 60462

Gordon's solution can be simplified to a single OLAP-function:

select min(datetime), max(datetime), count(*) as numrecs, avg(temperature)
from
 (
   select datetime, temperature, 
      -- previous time when temperature was < 50
      -- same time for all rows with a temp >= 50
      max(case when temperature < 50 then datetime end)
      over (order by datetime
            rows unbounded preceding) as prevlow
   from measurement
 ) as dt
where temperature >= 50
group by prevlow
having max(datetime) >= min(datetime) + interval '30' minute;

Upvotes: 3

Gordon Linoff
Gordon Linoff

Reputation: 1269773

So, first identity groups where the temperature is above 50. This is a gaps-and-islands problem. You can then summarize the island to get the information you want:

select min(datetime), max(datetime), count(*) as numrecs, avg(temperature)
from (select t.*,
             row_number() over (order by datetime) as seqnum,
             row_number() over (partition by (temperature >= 50)::int 
                                order by datetime) as seqnum_t
      from t
     ) t
where temperature >= 50
group by (seqnum - seqnum_t)
having max(datetime) >= min(datetime) + interval '30' minute;

Upvotes: 5

Related Questions