The Quantum approximate optimization algorithm is a quantum-classical hybrid algorithm, which can obtain the optimal solution of combinatorial optimization problems in polynomial time.But at a low level of iteration, there is no guarantee that the problem will be solved optimally. A quantum circuit with fewer quantum gates based on the revised target Hamiltonian was developed by this study to address this difficulty,the solution procedure was streamlined and solution precision was boosted. In order to verified the reliability of the proposed solution,experiments were carried out by solving the integer linear programming problem.The experiment was deployed in the Pyqpanda environment of the origin quantum. The findings show that the average execution time is 20.8% of the original, and the probability is enhanced from 54.156 3% to 82.9%.
近年来,物联网、移动边缘计算、车载自组织网络(vehicular ad hoc network,VANET)发展迅速,随之而来的一些基站和服务器选址、资源分配调度等问题受到广泛关注。这些问题可以理解为在有限的可供选择的方案中,寻找满足一定约束的最好方案,因此类似此类问题可以归结为整数线性规划问题[1]。整数线性规划问题属于线性规划中的一种,其中又分为二元整数线性规划、混合整数线性规划等。随着问题规模的扩大,在约束条件的限制下寻找到整数线性规划问题最优解的时间复杂度较高。
对于同一个问题,当设置的目标哈密顿量不同时,结果是不同的。在解决整数线性规划问题时,设置了两个不同的目标哈密顿量。一种是仅为问题函数(用 Hp 表示)求解目标哈密顿量;另一种是将问题函数和变形约束求和,并忽略公式中的常数,以获得改进的目标哈密顿量(用 Hc 表示)。以上两种方法用于设置不同的目标哈密顿量,为比较实验做准备。
本文利用QAOA解决了与资源分配相关的整数线性规划问题,将节点间的链路映射到量子位来编码哈密顿量 Hc 、 Hp 和 Hb。通过从原始的目标哈密顿量 Hp 转换到改进的目标哈密顿量 Hc,量子电路的运行时间和所需量子门的数量都显著减少。此外,本文分析并比较了原始目标哈密顿量 Hp 和改进的目标哈密顿量 Hc 对找到近似最优解的影响。实验结果证明了QAOA在解决这些问题方面的有效性,即使有少量的迭代,该算法也有很高的成功率。例如,当目标哈密顿量是 Hp 时,成功率是54.156 3%,而当它是 Hc 时,概率增加到82.9%。这种低迭代水平转化为实现算法所需的低电路深度,证实了其在近期量子机器上的可行性。由于QAOA的时间复杂性不受问题大小的影响,因此它更适合大规模和多节点场景。此外,哈密顿量 Hc 的电路需要更少的量子门,这可以缩短执行时间,平均执行时间为 Hp 电路的20.8%。在未来,将会对QAOA的内部结构进行改进,发现更有效的参数更新技术,并将不同的模型应用于不同的组合优化问题。
FarhiE, HarrowA W.Quantum supremacy through the quantum approximate optimization algorithm[J].Bulletin of the American Physical Society,2017,62(4):278-279.
[5]
WillschM, WillschD, JinF P,et al.Benchmarking the quantum approximate optimization algorithm[J].Quantum Information Processing,2020,19(7):197.
[6]
HerrmanR, TreffertL, OstrowskiJ,et al.Impact of graph structures for QAOA on MaxCut[J].Quantum Information Processing,2021,20(9):289.
[7]
BengtssonA, VikstålP, WarrenC,et al.Improved success probability with greater circuit depth for the quantum approximate optimization algorithm[J].Physical Review Applied,2020,14(3):034010.
[8]
BravyiS, KlieschA, KoenigR,et al.Obstacles to variational quantum optimization from symmetry protection[J].Physical Review Letters,2020,125(26):260505.
[9]
AzadU, BeheraB K, AhmedE A,et al.Solving vehicle routing problem using quantum approximate optimization algorithm[J].IEEE Transactions on Intelligent Transportation Systems,2023,24(7):7564-7573.
[10]
ZhangY J, MuX D, LiuX W,et al.Applying the quantum approximate optimization algorithm to the minimum vertex cover problem[J].Applied Soft Computing,2022,118:108554.
[11]
VikstlP, GrnkvistM, SvenssonM,et al.Applying the quantum approximate optimization algorithm to the tail⁃assignment problem[J].Physical Review Applied,2020,14(3):034009.
[12]
GongC Q, WangT, HeW Y,et al.A quantum approximate optimization algorithm for solving Hamilton path problem[J].The Journal of Supercomputing,2022,78(13):15381-15403.
[13]
ChoiJ, OhS, KimJ.Quantum approximation for wireless scheduling[J].Applied Sciences,2020,10(20):7116.
[14]
FanZ Q, XuJ C, ShuG Q,et al.Solving the shortest path problem with QAOA[J].SPIN,2023,13(1):2350002.
[15]
SolovievV P, BielzaC, LarranagaP.Quantum approximate optimization algorithm for Bayesian network structure learning[J].Quantum Information Processing,2022,22(1):1-28.
[16]
LucasA.Ising formulations of many NP problems[J].Frontiers in Physics,2014,2:5.
[17]
AharonovD, DamW V, KempeJ,et al.Adiabatic quantum computation is equivalent to standard quantum computation[J].SIAM Review,2008,50(4):755-787.
[18]
StreifM, LeibM.Training the quantum approximate optimization algorithm without access to a quantum processing unit[J].Quantum Science and Technology,2020,5(3):034008.
[19]
XiangT, WangJ, LiaoX F.An improved particle swarm optimizer with momentum[C]//2007 IEEE Congress on Evolutionary Computation.Singapore City,Singapore:IEEE,2008:3341-3345.
[20]
BabuD V, KarthikeyanC,Shreya,et al .Performance analysis of cost and accuracy for whale swarm and RMSprop optimizer[J].IOP Conference Series:Materials Science and Engineering,2020,993(1):012080.
[21]
GiannakasF, TroussasC, VoyiatzisI, et al. A deep learning classification framework for early prediction of team⁃based academic perfor⁃mance[J].Applied Soft Computing, 2021, 106:107355.
[22]
BockS, WeisM .A proof of local convergence for the Adam optimizer[C]//2019 International Joint Conference on Neural Networks (IJCNN).Budapest,Hungary:IEEE,2019:1-8.
[23]
MaC, WangK Z, ChiY J,et al.Implicit regularization in nonconvex statistical estimation:gradient descent converges linearly for phase retrieval,matrix completion,and blind deconvolution[J].Foundations of Computational Mathematics,2020,20(3):451-632.