MSSQL：识别时间序列中一系列不变（平线）值

Question

我正在处理作为传感器测量结果的时间序列数据。我需要确定数据平坦的情况 - 表明传感器故障。我想选择过去 24 小时内连续有 3 个以上不变的值。

我想我可能需要循环，但我没有在 sql 中使用循环。我假设我需要使用子查询来按日期时间排序。我也看过 LEAD 和 LAG。此外，我需要通过 SiteID 和 VariableID 来区分，我认为可以使用 PARTITION 来完成。

数据如下：

**SiteID**VariableID**DateTime**Value**
   5    1   2014-01-27 12:15    5.576
   5    1   2014-01-27 12:30    5.487
   5    1   2014-01-27 12:45    5.573
   5    1   2014-01-27 13:00    5.903
   5    87  2014-01-27 12:15    -273.2
   5    87  2014-01-27 12:30    -273.2
   5    87  2014-01-27 12:45    -273.2
   5    87  2014-01-27 13:00    -273.2
   5    88  2014-01-27 12:15    -273.2
   5    88  2014-01-27 12:30    -273.2
   5    88  2014-01-27 12:45    -273.2
   5    88  2014-01-27 13:00    -273.2
   5    89  2014-01-27 12:15    -273.2
   5    89  2014-01-27 12:30    -273.2
   5    89  2014-01-27 12:45    -273.2
   5    89  2014-01-27 13:00    -273.2
   5    2   2014-01-27 12:15    30.61
   5    2   2014-01-27 12:30    38.73
   5    2   2014-01-27 12:45    32.84
   5    2   2014-01-27 13:00    31.62
   5    3   2014-01-27 12:15    -9.53
   5    3   2014-01-27 12:30    -8.61
   5    3   2014-01-27 12:45    -8.76
   5    3   2014-01-27 13:00    -9.32
   5    4   2014-01-27 12:15    0.298
   5    4   2014-01-27 12:30    0.32
   5    4   2014-01-27 12:45    0.317
   5    4   2014-01-27 13:00    0.302

我想生成类似的东西：

**SiteID**VariableID**StartingDateTime**ValueCount**Value**
    5          87     2014-1-27 12:15       4         -273.4
    5          88     2014-1-27 12:15       4         -273.4
    5          89     2014-1-27 12:15       4         -273.4

Answer 1

SQL Fiddle

使用此架构和数据（稍作修改，以确保一切正常）：

CREATE TABLE TimeSeries (
  SiteId INT,
  VariableId INT,
  DateTime DATETIME,
  Value NUMERIC(15,5)
);

INSERT INTO TimeSeries VALUES (    5, 1   , '2014-01-27 12:15' ,   5.576     );
INSERT INTO TimeSeries VALUES (    5, 1   , '2014-01-27 12:30' ,   5.487     );
INSERT INTO TimeSeries VALUES (    5, 1   , '2014-01-27 12:45' ,   5.573     );
INSERT INTO TimeSeries VALUES (    5, 1   , '2014-01-27 13:00' ,   5.903     );
INSERT INTO TimeSeries VALUES (    5, 87  , '2014-01-27 12:15' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 87  , '2014-01-27 12:30' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 87  , '2014-01-27 12:45' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 87  , '2014-01-27 13:00' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 88  , '2014-01-27 12:15' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 88  , '2014-01-27 12:30' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 88  , '2014-01-27 12:45' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 88  , '2014-01-27 13:00' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 12:15' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 12:30' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 12:45' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 13:00' ,   -273.2    );
INSERT INTO TimeSeries VALUES (    5, 2   , '2014-01-27 12:15' ,   30.61     );
INSERT INTO TimeSeries VALUES (    5, 2   , '2014-01-27 12:30' ,   38.73     );
INSERT INTO TimeSeries VALUES (    5, 2   , '2014-01-27 12:45' ,   32.84     );
INSERT INTO TimeSeries VALUES (    5, 2   , '2014-01-27 13:00' ,   31.62     );
INSERT INTO TimeSeries VALUES (    5, 3   , '2014-01-27 12:15' ,   -9.53     );
INSERT INTO TimeSeries VALUES (    5, 3   , '2014-01-27 12:30' ,   -8.61     );
INSERT INTO TimeSeries VALUES (    5, 3   , '2014-01-27 12:45' ,   -8.76     );
INSERT INTO TimeSeries VALUES (    5, 3   , '2014-01-27 13:00' ,   -9.32     );
INSERT INTO TimeSeries VALUES (    5, 4   , '2014-01-27 12:15' ,   0.298     );
INSERT INTO TimeSeries VALUES (    5, 4   , '2014-01-27 12:30' ,   0.32      );
INSERT INTO TimeSeries VALUES (    5, 4   , '2014-01-27 12:45' ,   0.317     );
INSERT INTO TimeSeries VALUES (    5, 4   , '2014-01-27 13:00' ,   0.302     );

-- Just to make sure the query works
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 18:30' ,   10        );
INSERT INTO TimeSeries VALUES (    5, 89  , '2014-01-27 19:00' ,   -273.2    ); -- this is not a contiguous value

查询：

WITH Sequences AS (
  SELECT
    T.*,
    ROW_NUMBER() OVER (PARTITION BY SiteId, VariableId, Value ORDER BY DateTime) AS RNO,
    ROW_NUMBER() OVER (ORDER BY SiteId, VariableId, DateTime) AS RNE
  FROM
    TimeSeries T
)
SELECT
  S.SiteId,
  S.VariableId,
  S.Value,
  MIN(S.DateTime) AS [Start],
  MAX(S.DateTime) AS [End],
  COUNT(*) AS ValueCount
FROM
  Sequences S
GROUP BY
  S.SiteId,
  S.VariableId,
  S.Value,
  S.RNE - S.RNO
HAVING
  COUNT(*) > 1

Results:

| SITEID | VARIABLEID |  VALUE |                          START |                            END | VALUECOUNT |
|--------|------------|--------|--------------------------------|--------------------------------|------------|
|      5 |         87 | -273.2 | January, 27 2014 12:15:00+0000 | January, 27 2014 13:00:00+0000 |          4 |
|      5 |         88 | -273.2 | January, 27 2014 12:15:00+0000 | January, 27 2014 13:00:00+0000 |          4 |
|      5 |         89 | -273.2 | January, 27 2014 12:15:00+0000 | January, 27 2014 13:00:00+0000 |          4 |

可以看到，VariableId = 89 只找到了 4 条记录（因为不应该考虑我添加的最后 2 条记录）。

基于this SO answer 和this blog post。