基于改进神经网络的干点软测量

针对传统的神经网络收敛判断以模型的拟合精度为指标造成训练时间过长和过拟合等缺点,提出了一种改进神经网络(M-ANN).M-ANN将样本分成训练样本和校验样本,并提出了过拟合判据参数.通过训练样本采用误差反传算法对网络进行训练,训练过程中以模型对校验样本的预测性能为指标,通过过拟合判据参数的计算自适应地在获得具有最佳预测性能模型时终止网络训练.同时,针对影响初馏塔塔顶石脑油干点的因素众多且呈高度非线性的特征,应用M-ANN建立初顶石脑油干点软测量模型,获得模型的预测相对误差平方和均值比传统神经网络模型降低了27.5%.

Soft measurement of dry point based on modified artificial neural network

To overcome the two main flaws of traditional artificial neural network(ANN),i.e.over analogue and time-consuming training process,a novel modified artificial neural network(M-ANN) was proposed.The M-ANN divided a sample set into a training sample set and a testing sample set,and gave it an over analogue criterion of ANN.During the ANN training process based on the training sample set, the testing sample was applied to surveillance of the predicting ability of ANN.According to the calculated result of the over analogue criterion,the M-ANN was able to end ANN training process immediately after the optimal predicting ability model was obtained.Further,the M-ANN was employed to develop the model for soft measurement of the dry point of naphtha,with the consideration of the existing factors having effect on the naphtha dry point and the significant correlation among them.The predicting precision obtained by the M-ANN was 27.5% higher than the traditional ANN.