基于IPSO-Elman的气液两相流含气率测量方法

Void fraction measurement method of gas-liquid two-phase flow based on IPSO-Elman

为安全且非侵入式地测量气液两相流含气率,提出一种电阻层析成像(ERT)陈列电阻与Elman神经网络相结合的含气率测量方法。首先,为加快模型训练速度并避免数据冗余,使用主成分分析(PCA)算法对120维的阵列电阻特征降维。然后,在粒子群(PSO)算法中引入自适应惯性权重和非线性学习因子,并加入遗传算法(GA)的交叉和变异行为以加快算法收敛速度。最后,通过改进的粒子群(IPSO)算法优化Elman神经网络初始权值和阈值,并建立含气率测量模型。经对比实验发现,PCA-IPSO-Elman含气率测量模型的平均绝对百分比误差为2.92%,且训练时间较IPSO-Elman模型减少68.8%。说明所提方法可以达到预期的测量效果。

In order to obtain non-invasive measurement results of gas-liquid two-phase flow gas holdup safely,a method of measuring the void fraction based on electrical resistance tomography(ERT)array resistance value and Elman neural network was proposed.Firstly,in order to accelerate the training speed of model and avoid redundancy of data,principal component analysis(PCA)algorithm was used to reduce the dimension of the resistance feature of the 120-dimensional array.Then,the adaptive inertia weights and nonlinear learning factors were introduced into the particle swarm optimization(PSO)algorithm,and the crossover and mutation behaviors of genetic algorithm(GA)were added to improve the rate of convergence of the algorithm.Finally,the initial weights and thresholds of Elman neural network were optimized by improved particle swarm optimization(IPSO)algorithm,and gas holdup measurement model was established.Through the comparison experiment,it is found that the mean absolute percentage error of the gas holdup measurement model named PCA-IPSO-Elman is 2.92%and the training time of the model is reduced by 68.8%compared with IPSOElman model,which manifests that the proposed method can achieve the expected measurement effect.