论文部分内容阅读
采用变步长和惯性项调整权值系数及阔值的改进BP算法,建立了一润滑油调合BP神经网络系统。分别预测了内燃机稠化油100℃运动粘度(V100)和润滑油的配方组成。预测结果的误差分别为:内燃机稠化油V100的绝对误差(A.D.)在±0.1mm2/s范围内,相对误差(R.D.)在±1.5％范围内;润滑油调合各组分质量百分含量的绝对误差在±1.2％范围内,相对误差在±2.0％范围内。结果表明BP神经网络预测误差能满足试验要求,预测精度优于常规非线性回归方程(分别为:V100 A.D.在±0.5mm2/s范围内,R.D.在±4.0％范围内;组分含量A.D.在±3.0％范围内,R.D.在±4.0％范围内)。为润滑油调合和相关数据的计算提供了一种新方法。
An improved BP algorithm with variable steps and inertia to adjust the weight coefficient and threshold is used to build a BP neural network system. The composition of kinematic viscosity (V100) of 100 ℃ viscous oil of internal combustion engine and lubricating oil were predicted separately. The errors of the prediction results are as follows: the absolute error (AD) of the internal combustion engine thickened oil V100 is within ± 0.1mm2 / s and the relative error (RD) is within ± 1.5%; the mass percentages The absolute error is within ± 1.2% and the relative error is within ± 2.0%. The results show that the BP neural network prediction error can meet the test requirements, and the prediction accuracy is better than the conventional nonlinear regression equation (respectively: V100 AD within ± 0.5mm2 / s, RD within ± 4.0%; component content AD within ± 3.0%, RD within ± 4.0%). It provides a new method for the blending of lubricants and the calculation of related data.