量子粒子群算法及其在煤层气产能预测中的应用

发布时间：2018-04-16 07:47

本文选题：量子衍生粒子群优化 + 混合量子神经网络　；参考：《东北石油大学》2017年硕士论文

【摘要】：本文将煤层气的产能预测问题与粒子群优化相融合,首先提出了一种新的量子衍生粒子群优化算法,进而提出了一种混合量子衍生神经网络模型及算法。通过将上述两种算法相融合提出了基于量子粒子群优化的量子神经网络预测模型及算法,利用该模型逼近地质因素指标与产能间的非线性映射关系,以实现对煤层气产能精准预测的目的。论文具体研究成果如下。首先,本文提出了一种全新的量子衍生粒子群优化算法。粒子群优化(particle swarm optimization,PSO)自1995年提出以来,以其搜索效率高、收敛速度快以及控制参数少等特点,成功地吸引了广大学者的关注,对于该算法的改进也层出不穷。本文从个体的编码方式入手,提出了一种多比特概率幅编码的粒子群优化(multi-qubit probability amplitude PSO,MQPAPSO)算法。本算法的编码机制为,采用多比特量子系统的基态概率幅的方式进行编码。更新方法为采用张量积构造的多比特量子旋转门进行更新,其中旋转角度的确定取自自身最优粒子和全局最优粒子确定旋转角度。改进算法的优势在于只需更新一个量子比特的相位,即可更新该粒子的所有维,从而使优化能力明显提高。第二,通过将量子计算和神经计算相融合,提出了一种混合量子衍生神经网络模型(hybrid quantum inspired neural networ,HQINN)。其中隐层为量子衍生神经元,网络的量子权值的更新采用量子旋转门的方式。该模型可使权值调整更为精细,从而增强网络的映射能力。为进一步提高量子神经网络的全局收敛能力,本文提出基于MQPAPSO优化HQINN权值的新方法(MQPAPSO-HQINN),实验对比分析结果表明,MQPAPSO-HQINN的逼近能力及泛化能力都有较好的提升,从而验证了提出方法的有效性。最后,在理论算法的应用方面,选取山西省晋城的8口煤层气井的相关数据,利用改进后的神经网络(MQPAPSO-HQINN)建立煤层气产能预测模型,并与其他模型的预测结果进行对比,结果表明MQPAPSO-HQINN模型的预测精度与其他几种方法相比,结果最优。另外本文也对影响煤层气产能的5个地质参数进行了敏感性分析,并分析了不同参数对煤层气产能的影响程度。
[Abstract]:In this paper, the problem of coalbed methane productivity prediction is combined with particle swarm optimization. A new quantum derivative particle swarm optimization algorithm is proposed, and then a hybrid quantum derivative neural network model and algorithm are proposed.A quantum neural network prediction model and algorithm based on quantum particle swarm optimization (QPSO) is proposed by combining the above two algorithms. The model is used to approximate the nonlinear mapping relationship between geological factors and productivity.In order to achieve the goal of accurate prediction of coalbed methane production capacity.The specific research results are as follows.Firstly, a novel quantum derivative particle swarm optimization algorithm is proposed.Particle swarm optimization (PSO) has attracted the attention of many scholars because of its high search efficiency, fast convergence speed and less control parameters since it was proposed in 1995.In this paper, we propose a multi-qubit probability amplitude PSO-MQPAPSO algorithm for multi-bit probabilistic amplitude-coding based on individual coding methods.The coding mechanism of this algorithm is to code the ground state probability amplitude of multi-bit quantum system.The updating method is a multi-bit quantum rotary gate constructed by tensor product, in which the rotation angle is determined from its own optimal particle and the global optimal particle to determine the rotation angle.The advantage of the improved algorithm is that all the dimensions of the particle can be updated by updating the phase of a quantum bit, and the optimization ability is improved obviously.Secondly, a hybrid quantum inspired neural networn (HQINN) model is proposed by combining quantum computation with neural computing.The hidden layer is a quantum derived neuron, and the quantum weight of the network is updated by a quantum rotary gate.The model can make the weight adjustment more precise, thus enhancing the mapping ability of the network.In order to improve the global convergence ability of quantum neural networks, a new method of optimizing HQINN weights based on MQPAPSO is proposed in this paper. The experimental results show that the approximation ability and generalization ability of MQPAP-HQINN are better than those of MQPAPSO-HQINN.The validity of the proposed method is verified.Finally, in the application of theoretical algorithm, 8 coalbed methane wells in Jincheng, Shanxi Province are selected to establish the CBM productivity prediction model by using improved neural network MQPAPSO-HQINN, and the results are compared with those of other models.The results show that the prediction accuracy of MQPAPSO-HQINN model is the best compared with other methods.In addition, the sensitivity of five geological parameters affecting the productivity of coalbed methane is analyzed, and the degree of influence of different parameters on the productivity of coalbed methane is analyzed.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TE37;TP18

【相似文献】