• 对POI数据进行空间聚类，以高德学校数据为例，下图为数据空间分布：

  
  
  image
• 以下是聚类的Python代码，获得20个空间聚类中心。

from sklearn.cluster import KMeans
from sklearn.externals import joblib
import numpy
import matplotlib.pyplot as plt
import psycopg2


# -*- coding: utf-8 -*-
from sklearn.cluster import KMeans
from sklearn.externals import joblib
import numpy
import time
import matplotlib.pyplot as plt
print ("step 1: load data...")
dataSet = []
conn = psycopg2.connect(database="superpower", user="postgres", password="123456",
                        host="localhost",
                        port="5432");
cur = conn.cursor();
sql="select st_x(st_transform(t.geom,3857)) x,st_y(st_transform(t.geom,3857)) y from gaode_poi_school t";
cur.execute(sql);
keyData = cur.fetchall();
print(keyData);
cur.close();
conn.close();
for point in keyData:
    dataSet.append([float(point[0]), float(point[1])])

clf = KMeans(n_clusters=20)  # 设定k  ！！！！！！！！！！这里就是调用KMeans算法
s = clf.fit(dataSet)  # 加载数据集合
numSamples = len(dataSet)
centroids = clf.cluster_centers_
for center in centroids:
    conn = psycopg2.connect(database="superpower", user="postgres", password="123456",
                            host="localhost",
                            port="5432");
    cur = conn.cursor();
    print(center[0]);
    x=str(center[0]);
    y=str(center[1]);
    sql = "INSERT INTO centrois ( x, y) VALUES ('"+x+"', '"+y+"');";
    print(sql);
    cur.execute(sql);
    conn.commit();
    cur.close();
    conn.close();

下图为计算的聚类中心。

image

• 对POI进行分类：

--将中心点集转成geom，使用st_closestpoint 进行分类
INSERT into poi_class SELECT
    kl.geom,a.type center_type
FROM
    (
        SELECT
            st_closestpoint (
                ST_Collect (
                    ARRAY (SELECT T .geom FROM centrois T)
                ),
                K .geom
            ),
            K .*
        FROM
            gaode_poi_school K
    ) kl
INNER JOIN centrois A ON kl.st_closestpoint=a.geom

• 分类结果如下图：

image