SQL Data to interpolate/extrapolate - sql

Question

If i have a table that keeps a running average of kW usage at a certain temperature, and I wanted to get a kW usage for a temperature that has not been recorded before, how could i get either
(A) Two data points above or two points below the temperature to extrapolate.
(B) Closest data above and below the temperature to interpolate
The table temperatures looks like this
Column | Type | Modifiers | Storage | Stats target | Description
-------------------------+------------------+-----------+---------+--------------+---------------
temperature_io_id | integer | not null | plain | |
temperature_station_id | integer | not null | plain | |
temperature_value | integer | not null | plain | | in Fahrenheit
temperature_current_kw | double precision | not null | plain | |
temperature_value_added | integer | default 1 | plain | |
temperature_kw_year_1 | double precision | default 0 | plain | |
"temperatures_pkey" PRIMARY KEY, btree (temperature_io_id, temperature_station_id, temperature_value)
(A) Proposed Solution
This would be a bit easier I think. The query would order the rows by the temperature value > or < the temperature im going for, then limit the results to 2? This would give me the two closest values that are above or below the temperature. Of course the order would have to be descending and ascending to make sure i get the right side of the items.
SELECT * FROM temperatures
WHERE
temperature_value > ACTUALTEMP and temperature_io_id = ACTUAL_IO_id
ORDER BY
temperature_value
LIMIT 2;
I think similar to above, but just limit it to 1 and do 2 queries, one for > and the other for <. I feel like this could be done better though?
Edit - Some sample data
temperature_io_id | temperature_station_id | temperature_value | temperature_current_kw | temperature_value_added | temperature_kw_year_1
-------------------+------------------------+-------------------+------------------------+-------------------------+-----------------------
18751 | 151 | 35 | 26.1 | 2 | 0
18752 | 151 | 35 | 30.5 | 2 | 0
18753 | 151 | 35 | 15.5 | 2 | 0
18754 | 151 | 35 | 12.8 | 2 | 0
18643 | 151 | 35 | 4.25 | 2 | 0
18644 | 151 | 35 | 22.15 | 2 | 0
18645 | 151 | 35 | 7.45 | 2 | 0
18646 | 151 | 35 | 7.5 | 2 | 0
18751 | 151 | 34 | 25.34 | 5 | 0
18752 | 151 | 34 | 30.54 | 5 | 0
18753 | 151 | 34 | 15.48 | 5 | 0
18754 | 151 | 34 | 13.08 | 5 | 0
18643 | 151 | 34 | 4.3 | 5 | 0
18644 | 151 | 34 | 22.44 | 5 | 0
18645 | 151 | 34 | 7.34 | 5 | 0
18646 | 151 | 34 | 7.54 | 5 | 0

Answer 1

You can get the nearest rows using:
select t.*
from temperatures t
order by abs(temperature_value - ACTUAL_TEMPERATURE) asc
limit 2
Or, a better idea in this case, is union:
(select t.*
from temperatures t
where temperature_value <= ACTUAL_TEMPERATURE
order by temperature_value desc
limit 1
) union
(select t.*
from temperatures t
where temperature_value >= ACTUAL_TEMPERATURE
order by temperature_value asc
limit 1
)
This version is better because it returns only one row if the temperature is in the table. This is a case where the UNION and duplicate removal is useful.
Next use conditional aggregation to get the information needed. This uses a short-cut, assuming that the kw increases with temperature:
select min(temperature_value) as mintv, max(temperature_value) as maxtv,
min(temperature_current_kw) as minck, max(temperature_current_kw) as maxck
from ((select t.*
from temperatures t
where temperature_value <= ACTUAL_TEMPERATURE
order by temperature_value desc
limit 1
) union
(select t.*
from temperatures t
where temperature_value >= ACTUAL_TEMPERATURE
order by temperature_value asc
limit 1
)
) t;
Finally, do some arithmetic to get the weighted average:
select (case when maxtv = mintv then minkw
else minkw + (ACTUAL_TEMPERATURE - mintv) * ((maxkw - minkw) / (maxtv - mintv))
end)
from (select min(temperature_value) as mintv, max(temperature_value) as maxtv,
min(temperature_current_kw) as minkw, max(temperature_current_kw) as maxkw
from ((select t.*
from temperatures t
where temperature_value <= ACTUAL_TEMPERATURE
order by temperature_value desc
limit 1
) union
(select t.*
from temperatures t
where temperature_value >= ACTUAL_TEMPERATURE
order by temperature_value asc
limit 1
)
) t
) t;