IIWPSO-PNN在化工过程故障诊断中的应用

摘要

概率神经网络(PNN)已成功应用于化工过程故障诊断.在概率神经网络中,平滑参数对网络性能有很大的影响,并且很难确定.因此,采用粒子群优化(PSO)算法,寻找最优平滑参数.针对粒子群优化算法中线性变化的惯性权重易使其陷入局部极值问题,采用非线性变化的惯性权重替代线性变化的惯性权重,并将其应用于改进惯性权重粒子群(IIWPSO)算法.将IIWPSO算法应用于概率神经网络中(即IIWPSO-PNN),使其自动搜索并寻找最优的平滑参数用于概率神经网络的训练和测试.与前人提出的线性变化惯性权重、两种非线性变化的惯性权重(分别记为w1、w2和w3)进行比较,将w1、w2和w3应用于PSO-PNN中(分别记为PSO-PNN1、PSO-PNN2和PSO-PNN3).最后将IIWPSO-PNN应用于田纳西-伊斯曼过程中,与PNN、PSO-PNN、PSO-PNN1、PSO-PNN2和PSO-PNN3网络进行比较.试验结果表明:IIWPSO-PNN在解决故障诊断问题时,识别率与收敛速度都有较大的提高.试验结果验证了IIWPSO-PNN算法应用于化工过程的可行性和有效性.%Probabilistic neural network has been successfully applied in fault diagnosis of chemical process.In probabilistic neural network,the smoothing parameter has great influence on its performance and it is difficult to determine the optimal value.Therefore,the particle swarm optimization (PSO) algorithm is used to seek for the optimal smoothing parameter.In PSO algorithm,the inertia weight of linear variation is easy to make the algorithm fall into local extremum,so the inertia weight of nonlinear change is used to replace the inertia weight of linear change and apply into the improved inertia weight particle swarm optimization algorithm (IIWPSO).Then,the IIWPSO is applied to probabilistic neural network,which can automatically search and find the optimal smoothing parameter to be used for training and testing the probabilistic neural network.Compared with the inertia weight of the linear change and two kinds of nonlinear inertia weight (named w1,w2 and w3),the w1,w2 and w3 are applied to PSO-PNN respectively,which are denoted as PSO-PNN1,PSO-PNN2 and PSO-PNN3.Finally,the IIWPSO-PNN network is applied in Tennessee Eastman process,and compared with PNN,PSO-PNN,PSO-PNN1,PSO-PNN2 and PSO-PNN3 networks.The test results indicate that IIWPSO-PNN network has higher recognition rate and convergence rate when addressing the fault diagnosis problem.The feasibility and effectiveness of the IIWPSO-PNN algorithm in chemical process are verified.