您可以使用高阶函数，如 transform 在最新版本的Spark如下。数据：

val df = Seq(
  ("112240386", "7435038894", 2, Array(4659, 53540, 7890)),
  ("113325994", "7940375640", 1, Array(7232, 7840, 8396)),
  ("223352476", "7765270324", 4, Array(999)),
  ("345936074", "7950076012", 1, Array(78650, 4829, 3000)),
).toDF("key_1", "key_2", "key_3", "ref_code")

Spark3.0.0+

df.withColumn("ref_code", transform($"ref_code", (x, i) => struct(x, i) ))

Spark>2.4

df.withColumn("ref_code", expr("transform(ref_code, (x,i) -> (x,i) )"))

Spark<2.4

val addIndex = udf((arr: Seq[Int]) => arr.zipWithIndex)
df.withColumn("ref_code", addIndex($"ref_code")).show(false)

输出：

+---------+----------+-----+----------------------------------+
|key_1    |key_2     |key_3|ref_code                          |
+---------+----------+-----+----------------------------------+
|112240386|7435038894|2    |[[4659, 0], [53540, 1], [7890, 2]]|
|113325994|7940375640|1    |[[7232, 0], [7840, 1], [8396, 2]] |
|223352476|7765270324|4    |[[999, 0]]                        |
|345936074|7950076012|1    |[[78650, 0], [4829, 1], [3000, 2]]|
+---------+----------+-----+----------------------------------+

这里有更多关于变换的例子