我们可以使用堆来解决这个问题，使用一个最小堆来存储当前的数组。我们还需要一个辅助堆，将每次从数组中删除的元素的成本添加到辅助堆中。然后，我们可以继续从堆顶开始执行删除操作，直到我们得到一个空数组。删除时，我们需要根据操作的类型计算删除成本。

以下是Python代码实现：

import heapq

def min_cost_to_empty_array(arr):
    heap = []
    for num in arr:
        heapq.heappush(heap, num)
    helper_heap = []
    cost = 0
    while len(heap) > 0:
        if len(heap) == 1:
            heapq.heappush(helper_heap, (0, heapq.heappop(heap)))
        else:
            first = heapq.heappop(heap)
            if len(heap) == 1:
                heapq.heappush(helper_heap, (0, first+heapq.heappop(heap)))
            else:
                middle = heapq.heappop(heap)
                last = heapq.heappop(heap)
                heapq.heappush(helper_heap, (0, first+middle+last))
                heapq.heappush(heap, middle)
        cost += helper_heap[0][1]
        heapq.heappop(helper_heap)
    return cost

arr = [1, 2, 3, 4, 5]
print(min_cost_to_empty_array(arr)) # 33