Airbus Quantum Computing Challenge 2020

Implementation of the solution to the Airbus Quantum Challenge 2020 aircraft loading problem proposed in Aircraft Loading Optimization -- QUBO models under multiple constraints.
The optimization process can be run on both a simulated annealer and real D-Wave solver.
The current implementation lacks an effective strategy to tune the coefficients so a lot of runs are required.

How to run

• Install requirements pip install -r requirements.txt
• [Optional] Install the D-Wave inspector dwave install inspector
• [Optional] Create an account at D-Wave Leap
• [Optional] Take your API key then run dwave config create
• Run python main.py

Optional steps are required only to use the QPU solvers.

Customize problem parameters

To change the aircraft data create an AircraftData object and a shear curve. Two functions are provided to create symmetric and asymmetric linear curves.

    shear_curve = get_linear_shear_curve(4, 26000)
    acft = AircraftData(num_positions=4, payloa_area_length=40, max_payload=8000, shear_curve=shear_curve, min_cg=-0.1, max_cg=0.2)

To change the problem requirements first define the container types and masses as lists.
Each point of the shear curve must correspond to a container position with the center elements representing the central position left and right shears.

    cont_types = np.array(['t1', 't1', 't1', 't1', 't1', 't1'])
    cont_masses = np.array([2134, 3455, 1866, 1699, 3500, 3332])

Define the penalty functions coefficients as a dict

    coefs = {'pl_o': 1.0,
             'pl_w': 1.0,
             'pl_d': 1.0,
             'pl_c': 1.0,
             'cl_u': 1.0,
             'cl_l': 1.0,
             'cl_t': 1.0,
             'sl_l': 1.0,
             'sl_r': 1.0}

Finally create a new LoadingProblem and set its coefficients

    problem = LoadingProblem(acft, cont_types, cont_masses, 0.1, 120000, -0.05)
    problem.coefficients = coefs

Constraints can be enabled or disabled individually

    problem.disable_constraints({'sl'})
    problem.enable_constraints({'sl'})

By default only the payload constraints are enabled

Two methods are provided to tune the coefficients automatically for a give problem.

    problem.coefficients = tune_coefs_average(problem)

This method uses the average values of each penalty BQM

    problem.coefficients = tune_coefs_iterative(problem)

This method increases the coefficients iteratively until the penalty and objective functions have the same order of magnitude. The increase step can be customized.

The LoadingProblem class allows to get the BQM representation of each constraint as well as the total problem to pass to the solver. For example to create a BQM to pass to the solvers use

    bqm = problem.get_bqm()

The plot functions assume feasible solutions so first parse the solver results then remove the unfeasible solutoins. The plot the top k solutions.

    cont_occ_solutions = problem.parse_solution(result)
    cont_occ_solutions = problem.filter_solutions(cont_occ_solutions)

    plot_top_solutions(problem, cont_occ_solutions, 1)

Plots will be saved in the out directory.

Todo

• Payload constraints
• CG constraints
• Shear constraints
• Improve coefficient tuning. This should increase the ratio of feasible solutions found by the solvers
• Try hybrid solvers