量子混合蛙跳算法在过程神经网络优化中的应用

Application of Quantum Shuffled Frog Leaping Algorithm in Process Neural Networks Optimization

针对基于正交基展开的过程神经元网络参数较多,基函数展开项数和网络结构难以确定,传统BP算法不易收敛的问题,结合量子理论提出一种量子混合蛙跳算法,用于过程神经元网络的训练。该算法利用量子位的Bloch球面坐标将网络结构、网络参数和展开项数统一编码,提出沿球面上经过两点间的劣弧路径进行旋转的方法来同时更新三个优化解,并利用Hadamard门完成个体变异避免早熟,进而有效扩展解空间的搜索范围。以抽油机故障诊断和网络流量预测为例,验证了算法的有效性。

Aiming at the problems that there are many parameters in the process neural networks based on orthogonal ba-sis expansion, it is difficult to determine the basis function expansion items and network structure, and the traditional BP algorithm is difficult to converge. A quantum shuffled frog leaping algorithm is presented based on the quantum theory and is applied to train the process neural network. The network structure, network parameters and expand the number of items are unified encoded with Bloch spherical coordinates of qubits. The three optimal solution is updated by this method which the rotation is realized through along the spherical surface after minor path between two points. The mutation of individuals is completed with Hadamard gates, and then the search range of solution space is effectively extended. The effectiveness of the algorithm was proved by pumping unit fault diagnosis and network traffic prediction.