基于PLS的Elman神经网络算法研究

Elman Neural Network Algorithm Based on PLS

针对特征变量多的小样本，结合偏最小二乘（Partial Least Squares,PLS）法则原理与Elman神经网络结构性质，提出基于PLS的Elman神经网络算法（PLSElman）.新算法通过PLS对高维小样本进行特征降维时，顾及了与因变量的相关程度，所得到的数据进行网络训练和仿真，明显的简化了网络结构，且可得较精确的网络模型.通过实例分析，结果表明新算法提高了网络的收敛速度、预测的精准率，证明新算法提高网络处理问题的效率.同时为便于验证新算法的有效性，与基于主成分分析（Principal Component Analys，PCA）的Elman神经网络算法（PCAElman）进行了比较，PLSElman算法有明显的优越性.

As to small size samples which have many characteristic variables, when Partial Least Squares （PLS） principle and structural properties of Elman neural network are taken into account, PLS-Eiman is put forward. The new algorithm, when carry- ing feature reduction on high-dimensional and small size sample, takes its relativity to dependent variable into account. Obtained data carries on network training and simttlation, clearly simplifies network structure and can get more precise network models. According to case analysis, the result shows that new algorithm improves convergence rate of the network, the predicting precision and proves that new algorithm improves the efficiency of the dealing with problems of the network. In the meantime,in order to test the effec- tiveness of new algorithm, it is compared with Elman neural network algorithm based on Principal Component Analysis （PCA-E1- man） and it is observed that PLS-Elman algorithra has more advantages.