PostgreSQL PostGIS 的5种空间距离排序(knn)算法

摘要： 标签 PostgreSQL , PostGIS , operator , ops , knn 背景 PostgreSQL GiST索引支持排序接口，可以支撑空间、标准、数组、文本向量、文本等类型的排序。

标签

PostgreSQL , PostGIS , operator , ops , knn

---

背景

PostgreSQL GiST索引支持排序接口，可以支撑空间、标准、数组、文本向量、文本等类型的排序。

分别表示：

1、给定一空间位置，按空间距离排序输出

2、给定一数组、全文检索值、字符串，按数组、全文检索、字符串相似度排序输出

3、给定一标准类型的值，按标准类型的距离排序

数据库目前支持哪些排序操作符，可以参考：《PostgreSQL 如何确定某个opclass支持哪些操作符(支持索引)，JOIN方法，排序》

1. postgres=# select oprleft::regtype,oprright::regtype,oprresult::regtype,oprname,oprkind,oprcode from pg_operator where oid in (select amopopr from pg_amop where amopsortfamily<>0);

2. oprleft | oprright | oprresult | oprname | oprkind | oprcode

3. -----------------------------+-----------------------------+------------------+---------+---------+------------------------------------

4. point | point | double precision | <-> | b | point_distance

5. circle | point | double precision | <-> | b | dist_cpoint

6. polygon | point | double precision | <-> | b | dist_polyp

7. geometry | geometry | double precision | <-> | b | geometry_distance_centroid

8. geometry | geometry | double precision | <#> | b | geometry_distance_box

9. geometry | geometry | double precision | <<->> | b | geometry_distance_centroid_nd

10. geometry | geometry | double precision | |=| | b | geometry_distance_cpa

11. geography | geography | double precision | <-> | b | geography_distance_knn

12.  

13. oprcode是这个操作符对应的函数，要知道操作符的算法，可以下载postgis源码根据函数找到其C代码。

14.  

安装了PostGIS后，对于空间排序，目前支持5种算法

PostGIS 空间排序算法

1、2D平面，两个对象实际距离

<-> — Returns the 2D distance between A and B.    

2、计算两条有效轨迹类型的距离（如果两个轨迹没有交集, 返回空）

|=| — Returns the distance between A and B trajectories at their closest point of approach.  

3、2D平面，两个对象bound box实面的距离(所以如果BOUND BOX与另一对象 相交，包含，说明距离为0)

<#> — Returns the 2D distance between A and B bounding boxes.    

4、与1类似，多维对象，返回两个bound box 中心点的距离

<<->> — Returns the n-D distance between the centroids of A and B bounding boxes.  

5、与3类似，多维对象，返回两个bound box 实体的距离

<<#>> — Returns the n-D distance between A and B bounding boxes.  

例子

1、点和线段的实际距离

1. postgres=# select st_setsrid(st_makepoint(1,2),4326) <-> st_setsrid(ST_MakeLine(ST_MakePoint(0,0.9),

2. ST_MakePoint(3,4)),4326); ?column?

3. --------------------

4. 0.0463614285010702

5. (1 row)

2、点和线段（BOUND BOX实体）的距离

1. postgres=# select st_setsrid(st_makepoint(1,2),4326) <#> st_setsrid(ST_MakeLine(ST_MakePoint(0,0.9),

2. ST_MakePoint(3,4)),4326); ?column?

3. ----------

4. 0

5. (1 row)

3、两个不相交轨迹（范围没有相交）的距离

1. postgres=# WITH inp AS ( SELECT

2. ST_AddMeasure('LINESTRING Z (0 0 0, 10 0 5)'::geometry,

3. extract(epoch from '2015-05-26 10:01'::timestamptz),

4. extract(epoch from '2015-05-26 11:01'::timestamptz)

5. ) a,

6. ST_AddMeasure('LINESTRING Z (0 2 10, 12 1 2)'::geometry,

7. extract(epoch from '2015-05-25 01:00'::timestamptz),

8. extract(epoch from '2015-05-25 13:00'::timestamptz)

9. ) b

10. )

11. SELECT ST_DistanceCPA(a,b) distance FROM inp;

12. distance

13. ----------

14.  

15. (1 row)

4、两个有相交（范围有交集）的距离 (lower1upper1 与 lower2upper2 有相交)

1. postgres=# WITH inp AS ( SELECT

2. ST_AddMeasure('LINESTRING Z (0 0 0, 10 0 5)'::geometry,

3. extract(epoch from '2015-05-26 10:01'::timestamptz), -- lower

4. extract(epoch from '2015-05-26 11:01'::timestamptz) -- upper

5. ) a,

6. ST_AddMeasure('LINESTRING Z (0 2 10, 12 1 2)'::geometry,

7. extract(epoch from '2015-05-25 01:00'::timestamptz), -- lower

8. extract(epoch from '2015-05-26 13:00'::timestamptz) -- upper

9. ) b

10. )

11. SELECT ST_DistanceCPA(a,b) distance FROM inp;

12. distance

13. ------------------

14. 3.07479618495659

15. (1 row)

5、两个有相交（范围有交集）的距离

1. postgres=# WITH inp AS ( SELECT

2. ST_AddMeasure('LINESTRING Z (0 0 0, 10 0 5)'::geometry,

3. extract(epoch from '2015-05-26 10:01'::timestamptz),

4. extract(epoch from '2015-05-26 11:01'::timestamptz)

5. ) a,

6. ST_AddMeasure('LINESTRING Z (0 2 10, 12 1 2)'::geometry,

7. extract(epoch from '2015-05-25 01:00'::timestamptz),

8. extract(epoch from '2015-05-26 10:10'::timestamptz)

9. ) b

10. )

11. SELECT ST_DistanceCPA(a,b) distance FROM inp;

12. distance

13. -----------------

14. 10.621322893124

15. (1 row)

排序用法

select * from tbl order by geom <-> 某个空间对象常量;