时序SQL的常用的函数_云原生多模数据库 Lindorm(Lindorm)-阿里云帮助中心

本文介绍时序SQL中支持的常用函数和使用示例。

引擎与版本

常用时序函数仅适用于时序引擎。
时序引擎3.4.7及以上版本支持常用时序函数。如何查看或升级当前版本，请参见时序引擎版本说明和升级小版本。

常用时序函数列表

函数	说明
CURRENT_TIMESTAMP函数	获取系统当前运行时间。
SUM函数	计算时序数据表中Field列的和。
AVG函数	计算时序数据表中Field列的平均值。
COUNT函数	统计时序数据表中Field列值的个数。
MIN函数	查询时序数据表中Field列的最小值。
MAX函数	查询时序数据表中Field列的最大值。
FIRST函数	获取时序数据表中Field列的第一个值。
LAST函数	获取时序数据表中Field列的最后一个值。
PERCENTILE函数	计算时序数据表中与Field列关联的第P个百分位字段值。
RATE函数	计算时序数据表中Field列值的斜率。
DELTA函数	计算时序数据表中Field列的值与前一行对应值的差。
LATEST函数	获取时序数据表中Field列的最新的N条数据。
FORECAST函数（时序预测）	对时序数据表中的Field列进行时序预测。
ANOMALY_DETECT函数（时序异常检测）	对时序数据表中Field列进行时序异常检测。
BINS函数（特征分箱）	对时序数据表中Field列进行特征分箱。

数据准备

假设示例表sensor的结构如下：

+-------------+-----------+------------+
| columnName  | typeName  | columnKind |
+-------------+-----------+------------+
| device_id   | VARCHAR   | TAG        |
| region      | VARCHAR   | TAG        |
| time        | TIMESTAMP | TIMESTAMP  |
| temperature | DOUBLE    | FIELD      |
| humidity    | DOUBLE    | FIELD      |
+-------------+-----------+------------+

表中数据如下：

+-----------+----------+---------------------------+-------------+-----------+
| device_id |  region  |           time            | temperature | humidity  |
+-----------+----------+---------------------------+-------------+-----------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 12.100000   | 45.000000 |
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 13.200000   | 47.000000 |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   | 46.000000 |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 12.100000   | 51.000000 |
| F07A1261  | south-cn | 2021-04-22T15:51:21+08:00 | 13.200000   | 52.000000 |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | 10.600000   | 53.000000 |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 20.600000   | 55.000000 |
+-----------+----------+---------------------------+-------------+-----------+

共包含以下两条时间线：

时间线1：Tag为device_id=F07A1260, region=north-cn。Field为temperature、humidity。
时间线2：Tag为device_id=F07A1261, region=south-cn。Field为temperature、humidity。

说明

关于时间线的概念，请参见基本概念。

时间处理函数

CURRENT_TIMESTAMP函数

说明

获取系统当前运行时间。

语法

CURRENT_TIMESTAMP

重要

CURRENT_TIMESTAMP无参数，使用时直接通过关键字CURRENT_TIMESTAMP调用即可。
CURRENT_TIMESTAMP函数返回的时间戳精确至毫秒，例如2023-04-23T21:13:15.819+08:00。
CURRENT_TIMESTAMP函数返回的是UTC+8格式的Lindorm实例服务端的系统时间，使用该函数前请确认系统时间是否满足业务需求。
请尽量避免在实际业务写入数据时直接使用该函数的返回结果。建议在开发测试中需要大规模生成测试数据的场景下使用CURRENT_TIMESTAMP函数。

示例

在sensor表中插入一条数据，TAG为device_id=F07A1262, region=north-cn，时间戳为数据写入的当前时间。

INSERT INTO sensor(device_id,region,time,temperature,humidity) VALUES ('F07A1262','north-cn',CURRENT_TIMESTAMP,19.9,42);

聚合函数

SUM函数

说明

计算时序数据表中Field列的和。

语法

SUM(field_name)

说明

SUM的返回类型为DOUBLE或长整型BIGINT。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口计算temperature列的和。

SELECT device_id, region, time, sum(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 25.300000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 35.900000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，计算每个设备的温度之和。

SELECT device_id, region, time, sum(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 1970-01-01T08:00:00+08:00 | 35.900000   |
| F07A1261  | south-cn | 1970-01-01T08:00:00+08:00 | 56.500000   |
+-----------+----------+---------------------------+-------------+

示例3：计算所有设备的温度之和。

SELECT sum(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00';

返回结果如下：

+-------------+
| temperature |
+-------------+
| 92.400000   |
+-------------+

AVG函数

说明

计算时序数据表中Field列的平均值。

语法

AVG(field_name)

说明

AVG的返回类型为DOUBLE类型。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口计算temperature列的平均值。

SELECT device_id, region, time, avg(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 12.650000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 11.966667   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，计算每个设备的温度的平均值。

SELECT device_id, region, time, avg(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 1970-01-01T08:00:00+08:00 | 11.966667   |
| F07A1261  | south-cn | 1970-01-01T08:00:00+08:00 | 14.125000   |
+-----------+----------+---------------------------+-------------+

示例3：计算所有设备的温度的平均值。

SELECT avg(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00';

返回结果如下：

+-------------+
| temperature |
+-------------+
| 13.200000   |
+-------------+

COUNT函数

说明

统计时序数据表中Field列的值个数。

语法

COUNT(field_name)

说明

COUNT的返回类型为BIGINT类型。

参数

参数	描述
field_name	Field列名。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口统计各个窗口中temperature列的值个数。

SELECT device_id, region, time, count(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 2           |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 3           |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 1           |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 1           |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，统计每个设备的温度的值个数。

SELECT device_id, region, time, count(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 1970-01-01T08:00:00+08:00 | 3           |
| F07A1261  | south-cn | 1970-01-01T08:00:00+08:00 | 4           |
+-----------+----------+---------------------------+-------------+

示例3：统计所有设备的温度的值个数。

SELECT count(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00';

返回结果如下：

+-------------+
| temperature |
+-------------+
| 7           |
+-------------+

选择函数

MIN函数

说明

查询时序数据表中Field列的最小值。

语法

MIN(field_name)

说明

MIN的返回类型为DOUBLE或长整型BIGINT。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口查询各个窗口中temperature列的最小值。

SELECT device_id, region, time, min(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 12.100000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 10.600000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，查询每个设备的温度的最小值。

SELECT device_id, region, time, min(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | 10.600000   |
+-----------+----------+---------------------------+-------------+

示例3：查询所有设备的温度的最小值。

SELECT min(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00';

返回结果如下：

+-------------+
| temperature |
+-------------+
| 10.600000   |
+-------------+

MAX函数

说明

查询时序数据表中Field列的最大值。

语法

MAX(field_name)

说明

MAX的返回类型为DOUBLE或长整型BIGINT。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口查询各个窗口中temperature列的最大值。

SELECT device_id, region, time, max(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 13.200000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 13.200000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，查询每个设备的温度的最大值。

SELECT device_id, region, time, max(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 13.200000   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例三：查询所有设备的温度的最大值。

SELECT max(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00';

返回结果如下：

+-------------+
| temperature |
+-------------+
| 20.600000   |
+-------------+

FIRST函数

说明

获取时序数据表中Field列的第一个值。

语法

FIRST(field_name)

说明

FIRST的返回类型与Field列的类型相同。

参数

参数	描述
field_name	Field列名。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口获取各个窗口中第一个温度值。

SELECT device_id, region, time, first(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 12.100000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 12.100000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，获取每个设备第一个温度值。

SELECT device_id, region, time, first(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 12.100000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 12.100000   |
+-----------+----------+---------------------------+-------------+

LAST函数

说明

获取时序数据表中Field列的最后一个值。

语法

LAST(field_name)

说明

LAST的返回类型与Field列的类型相同。

参数

参数	描述
field_name	Field列名。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口获取各个窗口中最后一个温度值。

SELECT device_id, region, time, last(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 13.200000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 10.600000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，获取每个设备的最后一个温度值。

SELECT device_id, region, time, last(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

PERCENTILE函数

说明

计算时序数据表中与Field列关联的第P个百分位字段值。

语法

PERCENTILE(field_name,P)

说明

PERCENTILE的返回类型为DOUBLE类型。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。
P	取值为整数或浮点数，范围为[0,100]，默认值为50。

示例

示例1：设备ID为F07A1260和F07A1261，对每条时间线进行降采样操作，时间线按照20s时间窗口获取各个窗口中temperature列的90分位值。

SELECT device_id, region, time, percentile(temperature, 90) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00 | 13.200000   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00 | 13.200000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，获取每个设备的温度的90分位值。

SELECT device_id, region, time, percentile(temperature, 90) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 1970-01-01T08:00:00+08:00 | 13.200000   |
| F07A1261  | south-cn | 1970-01-01T08:00:00+08:00 | 20.600000   |
+-----------+----------+---------------------------+-------------+

转换函数

RATE函数

说明

计算时序数据表中Field列值的斜率。

计算公式：每个数据点的时间戳和值分别是：(t₁,v₁),(t₂,v₂)...(t_N,v_N)。使用RATE函数后，返回N-1个数据点，每个数据点的时间戳和值分别是：

2023-03-20_10-29-38

说明

公式中interval是指定的变化率，单位为秒。

语法

RATE(field_name, 'interval units')

说明

RATE的返回类型为DOUBLE类型。
SQL语句中只支持RATE算子单独使用。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。
interval units	变化率。默认值为1s。单位包括s（秒）、m（分钟）、h（小时）、d（天）。

示例

示例1：设备ID为F07A1260和F07A1261，计算每个设备温度的秒级变化率。

SELECT device_id, region, time, rate(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 0.110000    |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | -0.260000   |
| F07A1261  | south-cn | 2021-04-22T15:51:21+08:00 | 1.100000    |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | -0.433333   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 0.500000    |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，计算每个设备温度的分钟变化率。

SELECT device_id, region, time, rate(temperature, '1m') AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 6.600000    |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | -15.600000  |
| F07A1261  | south-cn | 2021-04-22T15:51:21+08:00 | 66.000000   |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | -26.000000  |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 30.000000   |
+-----------+----------+---------------------------+-------------+

示例3：设备ID为F07A1260和F07A1261，先按照20s时间窗口计算temperature列的平均值，再对降采样后的值计算分钟变化率。

SELECT device_id, region, time, rate(avg(temperature), '1m') AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | -6.150000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 25.900000   |
+-----------+----------+---------------------------+-------------+

示例4：SQL语句中只支持RATE算子单独使用，不支持以下混用方式。

错误示例1：

SELECT region, time, rate(temperature), rate(humidity) FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

ERROR 9001 (0A000): UNSUPPORTED: latest/rate/delta must be the only aggregator in a single SQL statement

错误示例2：

SELECT region, time, rate(temperature), latest(humidity) FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

ERROR 9001 (0A000): UNSUPPORTED: latest/rate/delta must be the only aggregator in a single SQL statement

DELTA函数

说明

计算时序数据表中Field列的值与前一行对应值的差。

计算公式：每个数据点的时间戳和值分别是：(t₁,v₁),(t₂,v₂)...(t_N,v_N)。使用DELTA函数后，返回N-1个数据点，每个数据点的时间戳和值分别是：(t₂,v₂-v₁),(t₃,v₃-v₂)...(t_N,v_N-,v_N-1)。

语法

DELTA(field_name)

说明

DELTA的返回类型为DOUBLE类型。
SQL语句中只支持DELTA算子单独使用。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。

示例

示例1：设备ID为F07A1260和F07A1261，计算每个设备温度与前一行差值。

SELECT device_id, region, time, delta(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 1.100000    |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | -2.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:21+08:00 | 1.100000    |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | -2.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 10.000000   |
+-----------+----------+---------------------------+-------------+

示例2：设备ID为F07A1260和F07A1261，先按照20s时间窗口计算temperature列的平均值，再对降采样后的值计算差值。

SELECT device_id, region, time, delta(avg(temperature)) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 20s;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | -2.050000   |
| F07A1261  | south-cn | 2021-04-22T15:51:40+08:00 | 8.633333    |
+-----------+----------+---------------------------+-------------+

示例3：SQL语句中只支持DELTA算子单独使用，不支持以下混用方式。

错误示例：

错误示例：SELECT region, time, delta(temperature), rate(humidity) FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

ERROR 9001 (0A000): UNSUPPORTED: latest/rate/delta must be the only aggregator in a single SQL statement

LATEST函数

说明

获取时序数据表中Field列的最新的N条数据。

语法

LATEST(field_name, N)

说明

LATEST的返回类型与Field列的类型相同。
SQL语句中只支持LATEST算子单独使用。

参数

参数	描述
field_name	Field列名。
N	整型，用于指定查询最新的N条数据。默认值为1，表示获取最新的一条数据。

示例

示例1：查询设备F07A1260和F07A1261最新的温度值。

SELECT device_id, region, time, latest(temperature) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 20.600000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
+-----------+----------+---------------------------+-------------+

示例2：查询设备F07A1260和F07A1261最新的两条温度值。

SELECT device_id, region, time, latest(temperature, 2) AS temperature FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+
| device_id |  region  |           time            | temperature |
+-----------+----------+---------------------------+-------------+
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00 | 10.600000   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00 | 20.600000   |
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00 | 13.200000   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00 | 10.600000   |
+-----------+----------+---------------------------+-------------+

示例3：SQL语句中只支持LATEST算子单独使用，不支持以下混用方式。

错误示例：

SELECT region, time, latest(temperature), rate(humidity) FROM sensor WHERE time <= '2021-04-22T15:51:47+08:00' and time >= '2021-04-22T15:33:00+08:00' sample by 0;

返回结果如下：

ERROR 9001 (0A000): UNSUPPORTED: latest/rate/delta must be the only aggregator in a single SQL statement

FORECAST函数（时序预测）

前提条件

时序预测功能依赖于Lindorm AI引擎，在使用FORECAST函数前，请开通Lindorm AI引擎。开通方式，请参见开通指南。

说明

对时序数据表中的Field列进行时序预测。

语法

FORECAST(field_name, model_name, options)

说明

FORECAST的返回类型为DOUBLE类型。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。
model_name	模型名称。说明 model_name的类型为VARCHAR类型。
options	调整时序预测的结果。可选参数。格式为：`key1=value1[, key2=value2]`。 batch_size：可选参数。一次推理所选取的时间线数量。当缓存的批量时间线数量达到batch_size时，执行一次批量预测。默认值为训练阶段时间线的数量，类型为INTEGER。 step：可选参数。输出预测结果的步长。默认值为训练阶段指定的PREDICTION_LENGTH参数的值，类型为INTEGER。 quantile_output：可选参数。输出分位数。对于时间线上的一个数据点，深度模型产生的预测结果是一个数组，通过指定quantile_output可以在数组中筛选数据作为最终预测结果。默认值为0.5，类型为DOUBLE。不同算法下取值不同：使用DeepAR算法时，取值范围为（0,1）。使用TFT算法时，取值为0.1，0.5或0.9。

示例

示例1：对时序数据表sensor中指定时间段的温度进行时序预测。

SELECT device_id, region, `time`, forecast(temperature, forecast_model) AS forecast_result FROM sensor WHERE `time` >= '2022-01-01T00:00:00+08:00' and `time` < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+------------------+
| device_id |  region  |           time            |  forecast_result |
+-----------+----------+---------------------------+------------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 |  12.40307807     |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 |  11.36715841     |
| F07A1260  | north-cn | 2022-01-01T00:00:40+08:00 |  10.12969923     |
| F07A1261  | south-cn | 2022-01-01T00:00:00+08:00 |  26.51369649     |
| F07A1261  | south-cn | 2022-01-01T00:00:20+08:00 |  25.54403301     |
| F07A1261  | south-cn | 2022-01-01T00:00:40+08:00 |  24.46405267     |
+-----------+----------+---------------------------+------------------+

示例2：对时序数据表sensor中指定时间段的温度进行时序预测，同时指定预测步长为2，输出分位数为0.9。

SELECT device_id, region, `time`, forecast(temperature, forecast_model, 'step=2,quantile_output=0.9') AS forecast_result FROM sensor WHERE `time` >= '2022-01-01T00:00:00+08:00' and `time` < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+------------------+
| device_id |  region  |           time            |  forecast_result |
+-----------+----------+---------------------------+------------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 |  13.12353792     |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 |  12.14833554     |
| F07A1261  | south-cn | 2022-01-01T00:00:00+08:00 |  26.73869304     |
| F07A1261  | south-cn | 2022-01-01T00:00:20+08:00 |  24.92990853     |
+-----------+----------+---------------------------+------------------+

ANOMALY_DETECT函数（时序异常检测）

说明

对时序数据表中Field列进行时序异常检测。

语法

ANOMALY_DETECT(field_name, [algo_name | model_name], options)

说明

ANOMALY_DETECT的返回类型为BOOLEAN类型。

参数

参数	描述
field_name	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。
algo_name	异常检测算法名称。支持阿里达摩院自研在线异常检测算法。 esd：适用于尖刺型异常（例如监控曲线出现尖刺的现象），对于数据点中有少量显著离群点的情况比较适用。 nsigma：原理简单，便于分析异常原因。 ttest：用于识别一个时间窗口内时序指标是否存在因均值变化而发生的异常。 istl-esd：适用于带周期性信号的数据。说明 algo_identifer参数适用于未开通数据库内机器学习功能，但有使用时序异常检测需求的场景。
model_name	模型名称。说明 model_name的类型为VARCHAR类型。 model_name参数仅适用于开通数据库内机器学习功能后，使用时序异常检测的场景。开通数据库内机器学习的方法，请参见开通方式。
options	调整异常检测算法的检测效果。可选参数。格式为：`key1=value1[, key2=value2]`。详细说明，请参见时序异常检测。

示例

示例1：对时序数据表sensor中指定时间范围的温度使用esd算法进行时序异常检测。

SELECT device_id, region, time, anomaly_detect(temperature, 'esd') AS detect_result FROM sensor WHERE time >= '2022-01-01T00:00:00+08:00' and time < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+---------------+
| device_id |  region  |           time            | detect_result |
+-----------+----------+---------------------------+---------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 | true          |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 | false         |
| F07A1260  | north-cn | 2022-01-01T00:00:40+08:00 | true          |
| F07A1261  | south-cn | 2022-01-01T00:00:00+08:00 | false         |
| F07A1261  | south-cn | 2022-01-01T00:00:20+08:00 | false         |
| F07A1261  | south-cn | 2022-01-01T00:00:40+08:00 | false         |
+-----------+----------+---------------------------+---------------+

示例2：对时序数据表sensor中F07A1260设备指定时间范围的温度使用esd算法进行时序异常检测。

SELECT device_id, region, time, anomaly_detect(temperature, 'esd') AS detect_result FROM sensor WHERE device_id in ('F07A1260') and time >= '2022-01-01T00:00:00+08:00' and time < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+---------------+
| device_id |  region  |           time            | detect_result |
+-----------+----------+---------------------------+---------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 | true          |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 | false         |
| F07A1260  | north-cn | 2022-01-01T00:00:40+08:00 | true          |
+-----------+----------+---------------------------+---------------+

示例3：对时序数据表sensor中F07A1260设备指定时间范围的温度使用esd算法进行时序异常检测，并自定义异常检测算法的参数（异常检测算法参考的时间窗口长度为30，异常检测时值小于90%分位数的点被过滤）。

SELECT device_id, region, time, anomaly_detect(temperature, 'esd', 'lenHistoryWindow=30,maxAnomalyRatio=0.1') AS detect_result FROM sensor where device_id in ('F07A1260') and time >= '2022-01-01T00:00:00+08:00' and time < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+---------------+
| device_id |  region  |           time            | detect_result |
+-----------+----------+---------------------------+---------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 | false         |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 | false         |
| F07A1260  | north-cn | 2022-01-01T00:00:40+08:00 | true          |
+-----------+----------+---------------------------+---------------+

示例4：使用在数据库内机器学习功能中创建的时序异常检测模型，对sensor表中指定时间范围的温度进行时序异常检测。

SELECT device_id, region, `time`, raw(temperature) as temperature, anomaly_detect(temperature, ad_model) AS detect_result FROM sensor WHERE time >= '2022-01-01T00:00:00+08:00' and time < '2022-01-01T00:01:00+08:00' sample by 0;

返回结果如下：

+-----------+----------+---------------------------+-------------+---------------+
| device_id |  region  |           time            | temperature | detect_result |
+-----------+----------+---------------------------+-------------+---------------+
| F07A1260  | north-cn | 2022-01-01T00:00:00+08:00 | 59.100000   | true          |
| F07A1260  | north-cn | 2022-01-01T00:00:20+08:00 | 13.200000   | false         |
| F07A1260  | north-cn | 2022-01-01T00:00:40+08:00 | 64.600000   | true          |
| F07A1261  | south-cn | 2022-01-01T00:00:00+08:00 | 12.100000   | false         |
| F07A1261  | south-cn | 2022-01-01T00:00:20+08:00 | 13.200000   | false         |
| F07A1261  | south-cn | 2022-01-01T00:00:40+08:00 | 10.600000   | false         |
+-----------+----------+---------------------------+-------------+---------------+

BINS函数（特征分箱）

说明

对时序数据表中Field列进行特征分箱，特征分箱（也称为离散分箱或者离散分段）是一种数据预处理技术。

语法

BINS(field_name, options)

说明

BINS的返回类型为VARCHAR类型。

参数

参数	是否必选	描述
field_name	是	Field列名。说明 Field列的类型不能是VARCHAR和BOOLEAN类型。
options	否	指定特征分箱策略和输出方式，具体说明请参见特征分箱。格式为：`key1=value1[, key2=value2]`。

示例

示例1：对时序数据表sensor中指定时间范围的温度进行特征分箱。

SELECT device_id, region, time, bins(temperature) AS temperature_bin FROM sensor WHERE time >= '2021-04-22 00:00:00' AND time < '2022-04-23 00:01:00' sample by 0;

返回结果如下：

+-----------+----------+--------------------------------+--------------------+
| device_id |  region  |           time                 |  temperature_bin   |
+-----------+----------+--------------------------------+--------------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00      |    [10.60,12.10)   |
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00      |    [12.10,13.20]   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00      |    [10.60,10.60)   |
| F07A1261  | south-cn | 2021-04-22T15:51:20+08:00      |    [10.60,12.10)   |
| F07A1261  | south-cn | 2021-04-22T15:51:21+08:00      |    [12.10,13.20)   |
| F07A1261  | south-cn | 2021-04-22T15:51:27+08:00      |    [10.60,10.60)   |
| F07A1261  | south-cn | 2021-04-22T15:51:47+08:00      |    [13.20,20.60]   |
+-----------+-----------+-------------------------------+--------------------+

示例2：对时序数据表sensor中F07A1260设备指定时间范围的温度进行特征分箱。

SELECT device_id, region, time, bins(temperature) AS temperature_bin FROM sensor WHERE device_id in ('F07A1260') AND time >= '2021-04-22 00:00:00' AND time < '2022-04-23 00:01:00' sample by 0;

返回结果如下：

+-----------+----------+--------------------------------+--------------------+
| device_id |  region  |           time                 |  temperature_bin   |
+-----------+----------+--------------------------------+--------------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00      |    [10.60,12.10)   |
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00      |    [12.10,13.20]   |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00      |    [10.60,10.60)   |
+-----------+----------+--------------------------------+--------------------+

示例3：对时序数据表sensor中F07A1260设备指定时间范围的温度进行特征分箱，并自定义分箱算法的参数（生成分箱数为2的等宽分箱，结果以分箱区间的顺序编码形式输出）。

SELECT device_id, region, time, bins(temperature, 'n_bins=2, output_type=ordinal, strategy=uniform') AS temperature_bin FROM sensor WHERE device_id in ('F07A1260') AND time >= '2021-04-22 00:00:00' AND time < '2022-04-23 00:01:00' sample by 0;

返回结果如下：

+-----------+----------+--------------------------------+--------------------+
| device_id |  region  |           time                 |  temperature_bin   |
+-----------+----------+--------------------------------+--------------------+
| F07A1260  | north-cn | 2021-04-22T15:33:00+08:00      |  1                 |
| F07A1260  | north-cn | 2021-04-22T15:33:10+08:00      |  1                 |
| F07A1260  | north-cn | 2021-04-22T15:33:20+08:00      |  0                 |
+-----------+----------+--------------------------------+--------------------+