数据分组根据在SQL Server

Question

的数据是15分钟的时间间隔相关的值：

 
Time    Value 
2010-01-01 00:15 3 
2010-01-01 00:30 2 
2010-01-01 00:45 4 
2010-01-01 01:00 5 
2010-01-01 01:15 1 
2010-01-01 01:30 3 
2010-01-01 01:45 4 
2010-01-01 02:00 12 
2010-01-01 02:15 13 
2010-01-01 02:30 12 
2010-01-01 02:45 14 
2010-01-01 03:00 15 
2010-01-01 03:15 3 
2010-01-01 03:30 2 
2010-01-01 03:45 3 
2010-01-01 04:00 5 
.......... 
.......... 
.......... 
2010-01-02 00:00 

通常，将存在96点。

根据这些值，我们可能注意到从00:15到01:45的值彼此接近，并且从02:00到03:00它们彼此接近，从03:15到04:00他们彼此接近。

基于 “相互靠近” 的规则，我想对数据进行 “分组” 3个部分：

• 00:15至01:45
• 02:00至03： 00
• 3点15到04:00

请考虑该数据可以是随机的，并且可以根据以上定义的规则被分成大于3份，但最大不应超过10部分。并且分组必须遵守时间顺序，例如，您不能仅将00：15/02：30/04：45分组为1组，因为这3个点不是连续的。

请介绍一下如何在t-sql中实现它。

更新： 的值可以是：

 
Time    Value 
2010-01-01 00:15 3 
2010-01-01 00:30 2 
2010-01-01 00:45 4 
2010-01-01 01:00 5 
2010-01-01 01:15 1 
2010-01-01 01:30 3 
2010-01-01 01:45 4 
2010-01-01 02:00 12 
2010-01-01 02:15 13 
2010-01-01 02:30 4 --suddenly decreased 
2010-01-01 02:45 14 
2010-01-01 03:00 15 
2010-01-01 03:15 3 
2010-01-01 03:30 2 
2010-01-01 03:45 3 
2010-01-01 04:00 5 
.......... 
.......... 
.......... 
2010-01-02 00:00 

对这类情况，我们不应该组单独02:30，因为我们想组的大小必须至少为3分，我们将把这一点（02:30）放到上一组（从02:00到03:00）。

Answer 1

由于您的问题改变了这么多，这里是一个新的答案，新问题，我只包含代码部分。

declare @t table(time datetime, value int) 
declare @variation float 
set @variation = 2 
set nocount on 

insert @t values('2010-01-01 00:15',3) 
insert @t values('2010-01-01 00:30',2) 
insert @t values('2010-01-01 00:45',4) 
insert @t values('2010-01-01 01:00',5) 
insert @t values('2010-01-01 01:15',1) 
insert @t values('2010-01-01 01:30',3) 
insert @t values('2010-01-01 01:45',4) 
insert @t values('2010-01-01 02:00',52) 
insert @t values('2010-01-01 02:15',5) 
insert @t values('2010-01-01 02:30',52) 
insert @t values('2010-01-01 02:45',54) 
insert @t values('2010-01-01 03:00',55) 
insert @t values('2010-01-01 03:15',3) 
insert @t values('2010-01-01 03:30',2) 
insert @t values('2010-01-01 03:45',3) 
insert @t values('2010-01-01 04:00',5) 


declare @result table(mintime datetime, maxtime datetime) 
a: 
delete @result 

;with t as 
(
select *, rn = row_number() over(order by time), log(value) lv from @t where datediff(day, time, '2010-01-01') = 0 
), a as 
(
select time, lv, rn, 0 grp from t where rn = 1 
union all 
select t1.time, a.lv, t1.rn, 
case when exists (select 1 from t t2 where t1.rn between rn + 1 and rn + 3 and 
lv between t1.lv - @variation and t1.lv [email protected]) then grp else grp + 1 end 
from t t1 join a on 
t1.rn = a.rn +1 
) 
insert @result 
select min(time), max(time) from a group by grp 

if @@rowcount > 10 
begin 
    set @[email protected] + .5 
    goto a 
end 

select * from @result 

结果：

mintime      maxtime 
2010-01-01 00:15:00.000  2010-01-01 01:45:00.000 
2010-01-01 02:00:00.000  2010-01-01 03:00:00.000 
2010-01-01 03:15:00.000  2010-01-01 04:00:00.000 

Answer 2

声明并填充TESTDATA：

set nocount on 
declare @result table(mintime datetime, maxtime datetime) 
declare @t table(time datetime, value int) 

-- variation is how much difference will be allowed from one row to the next 
declare @variation int 
set @variation = 5  

insert @t values('2010-01-01 00:15',3) 
insert @t values('2010-01-01 00:30',2) 
insert @t values('2010-01-01 00:45',4) 
insert @t values('2010-01-01 01:00',5) 
insert @t values('2010-01-01 01:15',1) 
insert @t values('2010-01-01 01:30',3) 
insert @t values('2010-01-01 01:45',4) 
insert @t values('2010-01-01 02:00',12) 
insert @t values('2010-01-01 02:15',13) 
insert @t values('2010-01-01 02:30',12) 
insert @t values('2010-01-01 02:45',14) 
insert @t values('2010-01-01 03:00',15) 
insert @t values('2010-01-01 03:15',3) 
insert @t values('2010-01-01 03:30',2) 
insert @t values('2010-01-01 03:45',3) 
insert @t values('2010-01-01 04:00',5) 

代码：

a: 

;with t as 
(-- add a rownumber 
select *, rn = row_number() over(order by time) from @t 
), a as 
(-- increase group if current row's value varies more than @variation from last row's value 
select time, value, rn, 0 grp from t where rn = 1 
union all 
select t.time, t.value, t.rn, case when t.value between 
     a.value - @variation and a.value [email protected] 
     then grp else grp+1 end 
from t join a on 
t.rn = a.rn +1 
) 
insert @result 
select min(time), max(time) from a group by grp 


if @@rowcount > 10 
begin 
    -- this will activate if more than 10 groups of numbers are found 
    -- start over with higher tolerance for variation 
    set @[email protected] + 1 
    delete @result 
    goto a 
end 

select convert(char(5), mintime,114) + ' to ' + convert(char(5), maxtime,114) 
from @result 

结果在这里： http://data.stackexchange.com/stackoverflow/q/110891/declare-and-populate-testdata