Question

我有关于包含位置ID和一组3 0或1个标志的位置的数据，这些标志指示位置的纬度，经度或地址是否已更改以及发生更改的月末

所以我看的是这样的事情：

+------------+-------------+--------------+---------------+---------------------+
| LOCATIONID | XCOORDHANGE | YCOORDCHANGE | ADDRESSCHANGE | REPORTPERIOD        |
+------------+-------------+--------------+---------------+---------------------+
| 1          | 0           | 0            | 1             | 2010-01-31 00:00:00 |
+------------+-------------+--------------+---------------+---------------------+
| 2          | 1           | 1            | 1             | 2010-03-31 00:00:00 |
+------------+-------------+--------------+---------------+---------------------+
| 1          | 1           | 1            | 0             | 2010-08-31 00:00:00 |
+------------+-------------+--------------+---------------+---------------------+

我的任务是确定移动的位置。移动被定义为x或y坐标变化和地址变化（有时位置被重新发现并且坐标发生变化但地址不会改变，有时地址会在没有后续坐标更改的情况下更改，我对此不感兴趣这些网站）。

确定何时将所有3个标志设置为1都很容易。问题是地址和坐标变化并不总是同时发生。例如，位置1显示2010年1月31日的地址更改，但是2010年8月31日的坐标更改。我需要查看每条记录，并确定在第一次更改后的一年内是否满足“移动”标准。对于上面示例中的位置1，如果x和/或y坐标的变化从地址变化开始长达1年（也就是说，标准在彼此的1年内得到满足），我认为它是“移动” 。添加的另一个皱纹是在我正在调查的4年期间内，位置可以移动多次。我在2010年1月31日到2014年12月31日这样做。

我的第一次尝试是使用ROW_NUMBER() OVER (PARTITION BY LOCATIONID ORDER BY REPORTPERIOD ASC) as rn并使用a.rn = a.rn+1上的自联接将一条记录链接到另一条记录，但这会忽略多次移动的位置。

最终目标是添加一个MEETSREQ列bit，其中1表示该位置有一个坐标更改和地址更改，这些更改发生在彼此的1年内。

输出看起来像这样

+------------+-------------+--------------+---------------+---------------------+---------+
| LOCATIONID | XCOORDHANGE | YCOORDCHANGE | ADDRESSCHANGE | REPORTPERIOD        | MEETREQ |
+------------+-------------+--------------+---------------+---------------------+---------+
| 1          | 0           | 0            | 1             | 2010-01-31 00:00:00 | 1       |
+------------+-------------+--------------+---------------+---------------------+---------+
| 2          | 1           | 1            | 1             | 2010-03-31 00:00:00 | 1       |
+------------+-------------+--------------+---------------+---------------------+---------+
| 1          | 1           | 1            | 0             | 2010-08-31 00:00:00 | 0       |
+------------+-------------+--------------+---------------+---------------------+---------+
| 3          | 0           | 0            | 1             | 2011-02-28 00:00:00 | 0       |
+------------+-------------+--------------+---------------+---------------------+---------+
| 4          | 1           | 1            | 0             | 2011-03-31 00:00:00 | 0       |
+------------+-------------+--------------+---------------+---------------------+---------+

这是SQL Server 2008 R2。感谢您的时间，我希望我已经添加了足够的清晰度。如有必要，我可以提供其他详细信息。

Answer 1

你可以这样做。注意虽然它是一个“邪恶”的光标。我个人觉得，当你执行复杂的业务逻辑时，它会保持清晰。

DECLARE @LOCATIONID INT
DECLARE @XCOORDHANGE INT
DECLARE @YCOORDCHANGE INT
DECLARE @ADDRESSCHANGE INT
DECLARE @REPORTPERIOD DATETIME


CREATE TABLE #Temp1 ( LOCATIONID INT, HASMOVED BIT );

-- find all locations that have an address change
DECLARE db_cursor CURSOR FOR  
SELECT LOCATIONID, XCOORDHANGE, YCOORDCHANGE, ADDRESSCHANGE, REPORTPERIOD 
FROM [TABLENAME]
WHERE ADDRESSCHANGE = 1

OPEN db_cursor   
FETCH NEXT FROM db_cursor INTO @LOCATIONID, @XCOORDHANGE, @YCOORDCHANGE, @ADDRESSCHANGE, @REPORTPERIOD 

WHILE @@FETCH_STATUS = 0   
BEGIN   

    -- find any other occurance of this location within the previous year, excluding any we've already looked at
    -- and must have an x or y coord change
    IF EXISTS(SELECT 0 FROM [TABLENAME] WHERE LOCATIONID = @LOCATIONID 
        AND LOCATIONID NOT IN(SELECT LOCATIONID FROM #Temp1)
        AND (XCOORDHANGE = 1 OR YCOORDCHANGE = 1)
        AND REPORTPERIOD > DATEADD(year, 1, @REPORTPERIOD)
    )
        INSERT INTO #Temp1 (LOCATIONID, HASMOVED) VALUES (@LOCATIONID, 1)
    ELSE
        INSERT INTO #Temp1 (LOCATIONID, HASMOVED) VALUES (@LOCATIONID, 0)

    FETCH NEXT FROM db_cursor INTO @LOCATIONID, @XCOORDHANGE, @YCOORDCHANGE, @ADDRESSCHANGE, @REPORTPERIOD    
END   

CLOSE db_cursor   
DEALLOCATE db_cursor

SELECT LOCATIONID, HASMOVED FROM #Temp1

如果您愿意，可以在现有的[TABLENAME]上加入现有的表格，包括HasMoved列。

这可能不是您指定的确切逻辑，但它应该为您提供我建议的方法的一般概念。

Answer 2

由于您只关心（x或y）和地址在一段时间内不是0，我在内部查询中使用了SUM

SELECT  LocationID
       ,SUM(Xcoord) AS x
       ,SUM(Ycoord) AS y
       ,SUM(Address) AS a
FROM myTable
WHERE Period BETWEEN '2010-01-01' AND '2010-12-31'
GROUP BY LocationID

然后在外部查询

上包含使用CASE的列

SELECT  LocationID
       ,(CASE WHEN (x > 0 OR y > 0) AND a > 0 THEN 1 ELSE 0 END) AS MeetsReq
FROM (
    SELECT  LocationID
           ,SUM(Xcoord) AS x
           ,SUM(Ycoord) AS y
           ,SUM(Address) AS a
    FROM myTable
    WHERE Period BETWEEN '2010-01-01' AND '2010-12-31'
    GROUP BY LocationID
) AS isrc

然后从基表中选择，左边加入子查询将MeetsReq的NULL值更改为0

/* This is the final query. 
   The 2 queries above are included here, 
   and was just separated for explanation purposes */

SELECT main.*, COALESCE(src.MeetsReq, 0) AS MeetsReq
FROM myTable AS main
LEFT OUTER JOIN (
    SELECT  LocationID
           ,(CASE WHEN (x > 0 OR y > 0) AND a > 0 THEN 1 ELSE 0 END) AS MeetsReq
    FROM (
        SELECT  LocationID
               ,SUM(Xcoord) AS x
               ,SUM(Ycoord) AS y
               ,SUM(Address) AS a
        FROM myTable
        WHERE Period BETWEEN '2010-01-01' AND '2010-12-31'
        GROUP BY LocationID
    ) AS isrc
) AS src ON main.LocationID = src.LocationID

虽然如果某个位置在MeetsReq上标记为1，那么该位置的所有记录都会相同。

Answer 3

在SQL Server 2012+中，您可以使用LEAD()执行此操作。

select t.*,
       (case when lead(addresschange) over (partition by locationid order by order by reportperiod) <> addresschange and
                  (lead(xcoordchange) over (partition by locationid order by reportperiod) <> xcoordchange or
                   lead(ycoordchange) over (partition by locationid order by reportperiod) <> ycoordchange
                  )
              then 0
              else 1
        end) as meetreq
from t;

在早期版本中，您可以使用outer apply：

select t.*
       (case when tnext.addresschange <> addresschange and
                  (tnext.xcoordchange <> xcoordchange or
                   tnext.ycoordchange <> ycoordchange
                  )
              then 0
              else 1
        end) as meetreq
from t outer apply
     (select top 1 t2.*
      from t t2
      where t2.locationid = t.locationid and t2.reportperiod > t.reportperiod
      order by t2.reportperiod asc
     ) tnext;

如何在一年内识别出某些标志等于1的记录？

3 个答案: