一天过去了 86400 几秒钟。你可以把它和epoch天相乘得到epoch秒，然后使用 from_unixtime .

scala> java.time.LocalDate.ofEpochDay(1)
res16: java.time.LocalDate = 1970-01-02

scala> java.time.LocalDate.ofEpochDay(2)
res17: java.time.LocalDate = 1970-01-03

scala> val df = Seq(1,2).toDF("days")
df: org.apache.spark.sql.DataFrame = [days: int]

scala> df.withColumn("date", from_unixtime(col("days")*86400, "yyyy-MM-dd").cast("date")).show
+----+----------+
|days|      date|
+----+----------+
|   1|1970-01-02|
|   2|1970-01-03|
+----+----------+

这是另一种方法。

df = spark.createDataFrame(
    [
        (1, 18508),

    ],
    schema=StructType(
        [
            StructField('Id', StringType(), True),
            StructField('Epoch_days', LongType(), True)
        ]
    )
)

df.withColumn(
    'date',
    (col('Epoch_days')*86400).cast('timestamp') 
).show(truncate=False)

# +---+----------+-------------------+

# |Id |Epoch_days|date               |

# +---+----------+-------------------+

# |1  |18508     |2020-09-03 00:00:00|

# +---+----------+-------------------+

val df = Seq(1).toDF("seq").select(
    from_unixtime(unix_timestamp(),"MM-dd-yyyy").as("date_1"),
    from_unixtime(unix_timestamp(),"dd-MM-yyyy HH:mm:ss").as("date_2"),
     from_unixtime(unix_timestamp(),"yyyy-MM-dd").as("date_3")
   ).show(false)

https://sparkbyexamples.com/spark/spark-epoch-time-to-timestamp-date/