基于粗糙集理论的PSO-IOIF-Elman神经网络建模  

PSO-IOIF-ELMAN NEURAL NETWORK MODELLING BASED ON ROUGH SET THEORY

作　　者：田娜[1] 南敬昌[1] 高明明[1] 

机构地区：[1]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105

出　　处：《计算机应用与软件》2016年第5期248-251,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(61372058);辽宁省高等学校优秀科技人才支持计划项目(LR2013012);辽宁工程技术大学研究生科研项目(5B2014032)

摘　　要：针对系统级仿真中功放建模的优缺点,结合粗糙集理论提出一种简化粒子群(PSO)算法优化改进OIF-Elman神经网络(PSO-IOIF-Elman)功放行为模型。该模型同时考虑小信号和大信号对功放记忆非线性的影响,结合AM-AM和AM-PM失真把OIFElman神经网络的自反馈系数用归一化后的输入输出电压表示。采用简化PSO优化算法,避免陷入局部最优,用粗糙集理论对模型预测值进行修正与补偿,提高预测精度。通过Matlab仿真比较,该模型训练误差减小9.53%,收敛速度提高11.31%,进而验证了建模方法的有效性和可靠性。In view of the advantages and disadvantages of power amplifier modelling in system-level simulation,this paper proposes a method using simplified particle swarm optimisation（ PSO） algorithm to optimise the improved OIF-Elman neural network（ PSO-IOIF-Elman）power amplifier behaviour model in combination with rough set theory. Considering different influences of small signal and large signal on the PA in regard to nonlinear characteristic of memory effect,and combing the characteristics of AM-AM and AM-PM modulation distortion,the model describes the self-feedback coefficient of OIF-Elman neural network to the normalised input and output voltage data. It employs the simplified PSO optimisation algorithm for preventing from falling into local optimal,and uses rough set theory to correct and compensate model＇s forecast value for improving the prediction precision. Through Matlab simulation comparison,the training error of the model reduces by 9. 53% and the convergence rate improves by 11. 31%,therefore verify the validity and reliability of the modelling method.

关 键 词：功放记忆非线性 行为模型 IOIF-Elman神经网络 简化粒子群算法 粗糙集理论 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]