基于量子蛙跳算法和过程神经网络的抽油机故障诊断

Pumping Unit Fault Diagnosis Based on Quantum Shuffled Frog Leaping Algorithm and Process Neural Networks

提出了一种量子混合蛙跳算法,该算法采用量子位的Bloch球面坐标编码个体,利用量子位在Bloch球面上绕轴旋转的方法实施优化搜索,采用Hadamard门实现个体变异以避免早熟,增强解空间的遍历性,可以快速逼近全局最优解。对过程神经网络的网络结构、网络参数和展开项数统一编码,并利用该算法进行优化,把优化后的神经网络应用到抽油机故障诊断中,结果表明,用量子混合蛙跳算法优化的神经网络对抽油机进行故障诊断较传统BP算法更具准确性与快速性。

A quantum shuffled frog leaping algorithm（QSFLA） was presented herein. In this algorithm, the individuals were expressed with Bloch spherical coordinate＇s of qubits, the evolution search was realized with the rotation of qubits in Bloch sphere. The mutation of individuals was achieved with Hadamard gates to avoid premature convergence. Above operations enhanced the ergodicity of the solution space and approximate global optimal solution fast. The network structure, network parameters and expand the number of items of PNN were encoded uniformly,and were optimized by the QSFLA. The optimizated neural network was used in pumping unit fault diagnosis. The diagnostic results between the new QSFLA and BP algorithm were compared. The conclusion is that the PNN based on QSFLA has better training performance,faster convergence rate and higher accuracy.