BPNN在改性石蜡滴熔点预测中的应用

针对传统误差反向传播神经网络(BPNN)的不足,在改进误差反向传播算法中引入了动量因子和非线性敏感度因子,实现了在学习过程中根据整体误差梯度变化对非线性敏感度因子进行动态调整。采用该BPNN模型对巴西棕榈蜡和川蜡改性的石蜡滴熔点进行了预测,预测结果的误差为改性石蜡滴熔点预测的绝对误差(A.D.)不超过±0.9℃,相对误差(R.D.)在±1.2%范围内。结果表明,改进的误差反向传播神经网络算法适用于改性石蜡滴熔点的预测,并具有较好的预测精度,可以减少石蜡调合试验及相应的改性石蜡滴熔点测试的次数,有利于提高工作效率,减少了资源浪费和环境污染。

Application of BPNN in Predicting Dropping-Melting Points of Modified Olefin

Momentum factor and nonlinear sensitivity factor was introduced in the modified back propagation neural network (BPNN) algorithm. The nonlinear sensitivity factor was adjusted according to the change of the integral deviation in the development. Dropping-melting points of the olefin modified by Brazil carnauba wax and Chinese wax were predicted using the modified BPNN. The predicting absolute deviations (A.D.) of dropping-melting points of the modified olefin are in the range of ±0.9℃ and the relative deviations (R.D.) are in the range of ±1.2%. The predicting results prove that the modified BPNN is applicable to the predicting the properties of the modified olefin. The predicting of dropping-melting points of the modified olefin gives the new way for the olefin modifying and the calculation of the relative data. This method decreases the times of the olefin modifying and the tests of dropping-melting points experiments, cuts down the expense of the manpower and material resources, decreases the waste of the natural resource and the pollution to the environment.