T-SQL时间平均

Question

我在SQL Server中有一个表来存储一块硬件的统计信息，表中的行表示给定秒的数据。它包含例如这些列：

timestamp (DateTime)
value (int)

我想要做的是从表格中选择给定日期/时间范围的数据，但是以给定时间段（例如1分钟，5分钟，1天等）的平均值返回在给定范围之间。所以一小时我有60行1分钟的平均值。

我从哪里开始？有人有什么意见或想法吗？

Answer 1

您可以在时间戳的DatePart上进行选择和分组。

例如：

SELECT
    DATEPART(hh, [timestamp]),
    DATEPART(mi, [timestamp]),
    AVG([value])
FROM
    YourTable
WHERE
    [timestamp] BETWEEN '2009-01-01 00:00:00.000' AND '2009-02-01 00:00:00.000'
GROUP BY
    DATEPART(hh, [timestamp]),
    DATEPART(mi, [timestamp])

编辑：对于更复杂的时间跨度，如5分钟，您可以按照以下方式对日期部分进行除法。

DATEPART(mi, [timestamp]) / 5 * 5

Answer 2

WITH    cal(m) AS
        (
        SELECT  1
        UNION ALL
        SELECT  m + 1
        FROM    cal
        WHERE   m < 60
        )
SELECT  DATEADD(minute, m, @start), AVG(value)
FROM    cal
LEFT JOIN
        timestamp
ON      timestamp > DATEADD(minute, m, @start)
        AND timestamp <= DATEADD(minute, m + 1, @start)
GROUP BY
        m

这将选择给定小时内所有分钟的平均值，即使是那些没有记录的分钟。

Answer 3

除了Robin Day的帖子之外，你可以按照5分钟的时间间隔进行分组：

GROUP BY
    DATEPART(hh, [timestamp]),
    DATEPART(mi, [timestamp]) / 5

如果您想延长几天，请参加dy，一年中的一天：

GROUP BY
    DATEPART(dy, [timestamp]),
    DATEPART(hh, [timestamp]),
    DATEPART(mi, [timestamp]) / 5

Answer 4

如果您要为此数据提供较高的读/写比率，则可能需要考虑索引视图。我已经在整个地方使用这种方法来按时间段聚合。我刚刚到了博客示例，这是代码：

create table timeSeries (
    timeSeriesId int identity primary key clustered
    ,updateDate datetime not null
    ,payload float not null
)

insert timeSeries values ('2009-06-16 12:00:00', rand())
insert timeSeries values ('2009-06-16 12:00:59', rand())
insert timeSeries values ('2009-06-16 12:01:00', rand())
insert timeSeries values ('2009-06-16 12:59:00', rand())
insert timeSeries values ('2009-06-16 01:00:00', rand())
insert timeSeries values ('2009-06-16 1:30:00', rand())
insert timeSeries values ('2009-06-16 23:59:00', rand())
insert timeSeries values ('2009-06-17 00:01:00', rand())
insert timeSeries values ('2009-06-17 00:01:30', rand())


create view timeSeriesByMinute_IV with schemabinding as
select
    dayBucket = datediff(day, 0, updateDate)
    ,minuteBucket = datediff(minute, 0, (updateDate - datediff(day, 0, updateDate)))
    ,payloadSum = sum(payLoad)
    ,numRows = count_big(*) 
from dbo.timeSeries
group by 
    datediff(day, 0, updateDate)
    ,datediff(minute, 0, (updateDate - datediff(day, 0, updateDate)))
go

create unique clustered index CU_timeSeriesByMinute_IV on timeSeriesByMinute_IV (dayBucket, minuteBucket)
go


create view timeSeriesByMinute as
select
    dayBucket
    ,minuteBucket
    ,payloadSum
    ,numRows
    ,payloadAvg = payloadSum / numRows
from dbo.timeSeriesByMinute_IV with (noexpand)
go

declare @timeLookup datetime, @dayBucket int, @minuteBucket int
select 
    @timeLookup = '2009-06-16 12:00:00'
    ,@dayBucket = datediff(day, 0, @timeLookup)
    ,@minuteBucket = datediff(minute, 0, (@timeLookup - datediff(day, 0, @timeLookup)))

select * from timeSeriesByMinute where dayBucket = @dayBucket and minuteBucket = @minuteBucket

您可以在代码块的末尾看到示例查找。显然，您可以定义要查询的范围，而不是仅仅寻找特定的dayBucket / minuteBucket对。

Answer 5

如果没有以下更改，我无法得到Quassnoi的工作答案：

WITH    cal(m) AS
    (
    SELECT  1
    UNION ALL
    SELECT  m + 1
    FROM    cal
    WHERE   m < 60
    )
SELECT  DATEADD(minute, m, @start) m, AVG(value)
FROM    cal
LEFT JOIN
    YourTable
ON      timestamp > DATEADD(minute, m, @start)
    AND timestamp <= DATEADD(minute, m + 1, @start)
GROUP BY
    m