一般安全的做法。是先將維一性條件及最新的一筆資料。先加入到一個temp表後。
再來做全數移除對應資料。
最後再從暫存檔中copy過來。

一般你說的方式要選擇的情況。太多多餘條件的情況下。很難給與一定符合的處理。
所以最好還是做分開動作會比較安全。

你試試..

select ID_PROJ
,NO_APPLY_A
,AM_APPLY
,UPDATE_USER
,UPDATE_TIME
from (
	select a.ID_PROJ
	,c.NO_APPLY_A
	,c.AM_APPLY
	,c.UPDATE_USER
	,c.UPDATE_TIME
	,ROW_NUMBER() Over (Partition By a.ID_PROJ,c.AM_APPLY Order by c.UPDATE_TIME desc) Sort
	from PRJECTA a
	left join PRJAPPLYA as c on a.id_proj=c.id_proj
	where a.DT_YYY='108' 
	and a.ID_DEPT_BUDG='505' 
	and c.DT_EXEC like '201901%' 
	and a.FG_SEAS<>'0'
	and c.UPDATE_USER='SYS'
) as k
where Sort = 1
order by UPDATE_USER DESC 
,UPDATE_TIME

以下是我自己try的..

declare @PRJECTA table(
	ID_PROJ varchar(50)
)

declare @PRJAPPLYA table(
	ID_PROJ varchar(50)
	,NO_APPLY_A varchar(50)
	,AM_APPLY money
	,UPDATE_USER varchar(50)
	,UPDATE_TIME time
)

insert into @PRJECTA
values('10850501100101')

insert into @PRJAPPLYA
values('10850501100101','a01',7673,'SYS','12:43:07')
,('10850501100101','a02',7673,'SYS','14:04:37')
,('10850501100101','a03',7719,'SYS','12:43:07')
,('10850501100101','a04',7719,'SYS','14:04:37')

select a.ID_PROJ
,c.NO_APPLY_A
,c.AM_APPLY
,c.UPDATE_USER
,c.UPDATE_TIME
,ROW_NUMBER() Over (Partition By a.ID_PROJ,c.AM_APPLY Order by c.UPDATE_TIME desc) Sort
from @PRJECTA a
left join @PRJAPPLYA as c on a.ID_PROJ=c.ID_PROJ
where c.UPDATE_USER='SYS'

select ID_PROJ
,NO_APPLY_A
,AM_APPLY
,UPDATE_USER
,UPDATE_TIME
from  (
	select a.ID_PROJ
	,c.NO_APPLY_A
	,c.AM_APPLY
	,c.UPDATE_USER
	,c.UPDATE_TIME
	,ROW_NUMBER() Over (Partition By a.ID_PROJ,c.AM_APPLY Order by c.UPDATE_TIME desc) Sort
	from @PRJECTA a
	left join @PRJAPPLYA as c on a.ID_PROJ=c.ID_PROJ
	where c.UPDATE_USER='SYS'
) as k
where Sort = 1
order by UPDATE_USER DESC 
,AM_APPLY
,UPDATE_TIME