算术错误导致cvxpy中的"Rank(A)<p或Rank([G; A])<n"错误

Question

我正在尝试使用cvxpy(以及cvxopt)在28个节点和37行的相对简单的网络中建模最优潮流,但得到"Rank(A)<p或Rank([G; A])<n "错误.

(使用相同的代码,我可以找到一个更简单的网络的最佳解决方案,具有4个节点和4个线路.)

我已经非常仔细地检查了约束是不一致的(使用我对Stack Overflow上这个问题的回答: 使用实际值检查cvxpy中的约束是否正常)

这是追溯:

File "main.py", line 201, in test_simple_optimisation
    p, bids, offers = optimize_flow(bids, offers, lines, nodes, injections_from_schedule, shift_factors, admittance.T)
  File "main.py", line 143, in optimize_flow
    p.solve()
  File "build/bdist.macosx-10.8-intel/egg/cvxpy/programs.py", line 169, in solve
obj,valid = solve_prog(new_p,quiet)
  File "build/bdist.macosx-10.8-intel/egg/cvxpy/procedures/solve_prog.py", line 44, in solve_prog
    sol = call_solver(p_expanded,quiet)
  File "build/bdist.macosx-10.8-intel/egg/cvxpy/procedures/call_solver.py", line 78, in call_solver
    r =  solvers.conelp(c,G,h,dims,A,b)
  File "/Library/Python/2.7/site-packages/cvxopt/coneprog.py", line 687, in conelp
    raise ValueError("Rank(A) < p or Rank([G; A]) < n")
ValueError: Rank(A) < p or Rank([G; A]) < n

当我查看/Library/Python/2.7/site-packages/cvxopt/coneprog.py",第687行时,它实际上是在拯救算术错误

 685         try: f = kktsolver(W)
 686         except ArithmeticError:..
 687             raise ValueError("Rank(A) < p or Rank([G; A]) < n")

那有意义吗？为什么会算术误差即类似的OverflowError,ZeroDivisionError,FloatingPointError,除非cvxpy已经扩展ArithmeticError意味着问题生病描述即,具有秩(A)<p或排名([G; A])<N"？

Answer 1

好吧,事实证明删除一些冗余约束确实解决了问题(正如cvxopt谷歌小组所建议的那样:https://groups.google.com/forum/？fromgroups =#!topic/cvxopt/qlu3CK1TdVQ .我不是确定为什么ArithmeticError被抛出.

实际上,我删除的约束不仅仅依赖于其他约束,而且保证满足任何输入,因为我是如何构造问题的.