按月份开始MS进行聚合，然后按loffset参数调整重新采样的时间标签：

df1 = df.resample('MS', loffset=pd.Timedelta(14, 'd')).sum()

样品：

rng = pd.date_range('2017-04-03', periods=15, freq='5D')
df = pd.DataFrame({'a': range(15)}, index=rng)  
print (df)
             a
2017-04-03   0
2017-04-08   1
2017-04-13   2
2017-04-18   3
2017-04-23   4
2017-04-28   5
2017-05-03   6
2017-05-08   7
2017-05-13   8
2017-05-18   9
2017-05-23  10
2017-05-28  11
2017-06-02  12
2017-06-07  13
2017-06-12  14

df1 = df.resample('MS', loffset=pd.Timedelta(14, 'd')).sum()
print (df1)
             a
2017-04-15  15
2017-05-15  51
2017-06-15  39

df1 = df.resample('SMS').sum()
print (df1)
             a
2017-04-01   3
2017-04-15  12
2017-05-01  21
2017-05-15  30
2017-06-01  39

另一个答案在pandas 1.4.2中不推荐使用，并带有警告FutureWarning: 'loffset' in .resample() and in Grouper() is deprecated.
推荐的替代方法是首先正常重采样，然后将Timedelta添加到索引：

df1 = df.resample('MS').sum()
df1.index += pd.Timedelta(14, 'd')

对于python 3.11和Pandas 2.0.2，下面的工作是在月中时间重新采样。

from datetime import datetime
times = ['2022-06-15 00:12:23', '2022-06-18', '2022-07-03', '2022-07-18']
data = [10, 100, 1000, 10000]
df = pd.DataFrame(data, index=pd.to_datetime(times), columns=['test'])

period = '1M'
start = pd.Timestamp(year=2022, month=5, day=15)
offset = pd.Timedelta(days=start.day - 1, hours=start.hour, minutes=start.minute, seconds=start.second, microseconds=start.microsecond)
df.index = df.index - offset
rs = df.resample(period).sum()
rs.index = rs.index + offset + pd.Timedelta(days=1)
rs

给了我：

test
2022-07-15  1110
2022-08-15  10000